Docstoc

Housing Wealth_ Liquidity Constraints_ and College Enrollment

Document Sample
Housing Wealth_ Liquidity Constraints_ and College Enrollment Powered By Docstoc
					       Housing Wealth, Liquidity Constraints, and
                 College Enrollment
                                       Michael F. Lovenheim ∗†
                                     SIEPR, Stanford University
                                             September 2008




                                                  Abstract
          A central empirical question in economics is whether liquidity constraints prohibit indi-
      viduals from making optimal consumption and investment decisions over the life cycle. I test
      for the existence of credit constraints surrounding the decision to invest in college, which is
      of particular concern given the large increases in the real cost of college attendance since the
      1970s. I add to the literature on higher education credit constraints by analyzing the role of
      housing wealth, which has not been examined previously. Using data from the Panel Study of
      Income Dynamics (PSID), I identify how changes in a household’s housing wealth in the 4 years
      prior to a child being of college-age affect the likelihood that child attends college. My results
      indicate that the 4-year growth in housing equity raises college enrollment and that the effect
      is localized to the 2000s and to households most likely to be credit constrained. Furthermore,
      I find that those living in cities with high housing price growth and low population growth are
      responsible for the entire estimated effect, which suggests endogenous housing price growth is
      not driving my results. The central implication of this work is college enrollment is sensitive
      to fluctuations in the housing market, which has important ramifications for the longer-run
      supply of high-skilled labor given the recent national slowdown in the housing market.

      KEYWORDS: Housing, Credit Constraints, Liquidity Constraints, Higher Education, College
      Enrollment.
      JEL CLASSIFICATION: E21, I22


   ∗ I would like to thank John Bound, Giacomo De Giorgi, Luigi Pistaferri, Nir Jaimovich, Caroline Hoxby, Lance

Lochner, Sarah Turner, John Pencavel, Greg Rosston, Raj Chetty, Manuel Amador and seminar participants at
Stanford University, the Institute for Research on Poverty Summer Workshop, and the American Education Finance
Association Annual Meeting for helpful comments. This research has been generously supported by the Searle
Freedom Trust and by the Stanford Institute for Economic Policy Research (SIEPR). All errors, omissions and
conclusions are my own.
   † Author contact information: Stanford Institute for Economic Policy Research, Stanford University, 579 Serra

Mall at Galvez Street, Stanford, CA 94305 ; email: mlovenhe@stanford.edu; phone: (650)736-8571.
1    Introduction

The existence and extent of credit constraints in the college enrollment decision has
received considerable attention in the economics literature. In the classical models
of education investment that assume perfect access to capital markets (Becker, 1964;
Ben-Porath, 1967; Mincer, 1997), individuals invest in higher education until their
internal rate of return equals the market rate of return to the investment. If individ-
uals are credit constrained, however, some potential students for whom the market
rate of return is sufficiently large to justify the investment may not go to college.
Thus, with incomplete capital markets, higher tuition and fees can lead to reduced
college enrollment even for those who have a positive lifetime return from investing
in a college education.
    The inability of households to finance an investment in higher education is eco-
nomically meaningful for two reasons. The first is it can restrict the supply of
college-educated, high skilled workers. There is evidence high skilled labor supply
fuels economic growth (DeLong, Goldin and Katz, 2003), so the lack of perfect credit
markets can have negative longer-run effects on the growth rate of the U.S. economy.
Second, depending on the joint distribution of the returns to education and family
resources, imperfect capital markets can cause an allocative inefficiency in who goes
to college.
    Because of the importance of supplying high-skilled labor to the workforce and
the growing cost of obtaining a college degree, there is a large literature on the ex-
istence of credit constraints and their relevance for educational attainment. Much
of this literature focuses on the positive correlation between income and schooling,
which is often taken as evidence of credit constraints (Ellwood and Kane, 2000). This
interpretation of the positive income gradient in collegiate attainment is confounded
by the strong association between student ability and family resources; controlling
for ability measures significantly reduces the enrollment gap across the income dis-



                                           1
tribution (Carneiro and Heckman, 2002; Cameron and Heckman, 1998 and 2001).1
For example, after controlling for AFQT quartile, Carneiro and Heckman (2002) find
only a weak relationship between income and educational attainment and argue only
8% of the U.S. population is credit constrained. In work using more recent data,
Belley and Lochner (2007) find a much stronger relationship between family income
and children’s educational attainment, suggesting parental resources are becoming
increasingly important in the higher education investment decision.
   In this paper, I contribute to the existing literature on credit constraints in higher
education by showing that housing wealth2 became an important component of the
college attendance decision over the past 10 years. My focus on the relationship
between college attendance and family housing wealth is novel to the literature, as
previous work has analyzed exclusively the relationship between college enrollment
and family income.3 I focus on the period before and during the millennial hous-
ing boom, which provided significant and plausibly exogenous variation in wealth to
homeowners. This paper represents an advancement over previous analyses because,
rather than assume family resources are exogenous conditional on family character-
istics and measured student ability, I am able to characterize more fully the wealth
variation I use to identify post-secondary education credit constraints. As a result,
my estimates can shed light on the existence and extent of credit constraints in higher
education.
   There are a number of reasons to expect housing wealth to be an important factor
in the college enrollment decision. First, 85% of college attendees come from families
who own a home, and housing wealth comprises the majority of household wealth
for most Americans. Thus, ignoring housing wealth will cause one to mismeasure
the extent of family resources, which can have significant effects on the estimation
   1 Similarly, Shea (2001) finds that while income is positively correlated with educational attainment in the raw

data, when he instruments for income using parental labor supply shocks, he finds no evidence of short-run family
income changes on education levels.
   2 Throughout this paper, I use the terms “housing wealth” and “housing equity” interchangeably.
   3 Belley and Lochner (2007) control for total family wealth quartile in attendance regressions using the NLSY97

and find a steep wealth gradient in college enrollment. However, as noted by the authors, their wealth variation is
probably endogenous, reflecting family decisions to save for their children’s college attendance.



                                                        2
of credit constraints. Second, the housing boom that occurred beginning in the late
1990s was characterized by both large increases in home values and an increasing
liquidity of accumulated home equity. For example, between 1990 and 2005, real
median home prices in the United States increased by 54%, and extracted home
equity as a proportion of household income increased by over 361%. In the 2000s,
a family that experienced a large increase in the value of its home around the time
period surrounding its child’s college attendance decision would have a significantly
easier time financing college expenditures due to the increased ease of borrowing
against its home’s value. The sharp housing price increases that characterized the
recent housing boom occurred differentially across both geographic areas and time,
which will allow me to identify those who experienced the largest home price increases
and determine whether their children were more likely to attend college.
   The analysis begins with a simple Mincer model of education investment that I
have augmented to incorporate the cost of funds. The insight of the canonical Mincer
model is that the optimal education investment for each individual is to attend college
if the discounted returns at the market rate of interest are greater than the market
rate of return. However, for many students, the cost of funds required to finance
college attendance may be higher than the discounted returns. Such students are
“credit constrained,” and insofar as an increase in the value of their parent’s home
reduces the cost of funds, I show that housing wealth variation can be used to identify
credit constraints in the higher education investment decision.
   I then test this hypothesis empirically using the Panel Study of Income Dynamics
(PSID), which contains a rich set of information about each child and each family,
including each family’s home value and accumulated home equity. I examine the
effect of the growth in each family’s home equity in the four years before a child is of
college-age between 1980 and 2005 on college attendance. Because there were large
increases in the both the real cost of college attendance over this time period as well as
the value and liquidity of housing, I compare effects across the 1980s, 1990s and 2000s.


                                            3
Consistent with the timing of college cost increases and the housing boom, my results
indicate housing wealth increases had no effect on enrollment in the 1980s and had
a positive but statistically insignificant effect in the 1990s. However, between 2000
and 2005, when housing wealth was the most liquid, I find an enrollment elasticity
with respect to housing wealth of 0.015, which is large enough to be economically
meaningful given the large fluctuations in real housing wealth over the sample period.
I calculate that about 6.8 percent of households use housing wealth to relax potential
credit constraints in college enrollment.
   Because the growth in housing prices may be correlated with local labor mar-
ket conditions, I split my sample into likely constrained and likely unconstrained
households, which are defined by income levels, home equity levels, and the ratio of
income to housing equity. In the 2000s, when housing wealth has the largest effect
on enrollment, I find effects are restricted to those groups who are more likely to be
credit constrained. I also show the positive relationship between housing wealth and
college enrollment is occuring solely in the MSAs that experienced high price growth
relative to population growth, which suggests that the housing wealth increases I use
to identify my empirical models is exogenous.
   Throughout this analysis, I maintain the assumption that college is an investment
good rather than a consumption good, which means the positive relationship between
housing wealth growth and college enrollment is indicative of credit constraints rather
than a wealth effect. I provide suggestive evidence in support of this assumption by
analyzing the effect of home equity changes on several consumption measures from
the PSID: total food expenditures, number of automobiles, and the number and
length of vacations. Consumption of these goods does not track the effect of housing
wealth on college enrollment, either temporally or geographically, which implies the
estimates are more in line with liquidity constraints rather than a general wealth
effect.
   My results have particularly serious policy implications given the recent slowdown


                                            4
in housing prices and the increasing difficulty of tapping one’s home equity.4 The
results from this analysis suggest the collapse of the housing bubble and the recent
“credit crunch” will reduce college enrollment by upwards of 4 percent and may thus
have negative longer-run effects on the supply of skilled labor in the United States.
     The rest of this paper is organized as follows: Section 2 presents a model of
education attainment that incorporates the cost of funds for college investment,
Section 3 discusses the PSID data used throughout this analysis, and Section 4
presents tabulations on housing wealth, income, and college enrollment. Section 5
describes the empirical models as well as the assumptions underlying identification
of credit constraints. Results are presented in Section 6, and Section 7 concludes.



2      Optimal Educational Investment and Liquidity Constraints

This section presents a theoretical model of education investment and outlines the
definition of credit constraints I will use throughout the analysis. I employ a simple
Mincer optimal education investment model augmented to take into account the fact
that agents face different interest rates from borrowing and lending, which largely
has been ignored in models of education investment. I follow Cameron and Taber
(2004) in assuming the market rate of return and the cost of funds may not be equal.5
Specifically, assume each individual, i, can borrow from one of N different assets at
interest rates r1 , . . . , rN . Without loss of generality, assume r1 <, . . . , < rN . One
of these interest rates is the “market interest rate,” rm , which is the real rate of
return on private savings that individuals can get in the marketplace with the same
risk as college investment. Thus, rm is the rate at which the forgone earnings from
attending college and the net present value of tuition payments should be evaluated.
     Following a Mincer-like setup, let the expected wage for a given college-goer be
    4 For
        example, in August 2007, National City Bank’s home equity unit froze all new applications for home equity
loans (Grant, 2007).
   5 While Cameron and Taber (2004) use this fact in their model, they do not focus on why this is so nor do they

examine how this difference changes with variation in family assets.




                                                       5
Yci and the expected wage for a non-college goer be Y0 . Note that Y0 is fixed for all
agents, but Yci varies across individuals.6 I assume the future wages of college-goers
can be described by an increasing function of ability, f (a), and a random component
described by the random variable ν with distribution G(ν). The expected college
wage is therefore Yci = f (ai ) + E(ν). The standard deviation of ν describes the risk
associated with college investment. At the extreme, if ν is zero, college investment
is riskless, and rm would equal the risk free rate of return. On the other hand, if
ν is the same as the standard deviation from a portfolio of average stock market
returns, rm will equal the average rate of return in the stock market. The small
literature on the riskiness of college investment suggests returns are high and the
standard deviation of returns is relatively low (Palacios-Huerta, 2003), which means
rm is probably between the risk-free rate of return and average stock market returns.
    The riskiness of college investment is also affected by the extent to which ν is cor-
related with the returns on other financial investments. The higher this correlation,
the higher is the aggregate risk taken on by those investing in college, which means
the opportunity cost of funds used to finance the college investment will increase due
to a higher rm .7
    The direct pecuniary cost of a year of college is p, which can be interpreted as the
cost of tuition, fees and books. I model education investment as a binary choice, with
each individual either investing in T years of education or not.8 For an individual
who invests in college, the cost of T years of education is the direct cost of education
   6 This model embodies the assumption that expected post-college earnings are a constant for each agent. While

this is a strong assumption, what matters for the optimal investment decision is the net present value of post-collegiate
earnings. Fixing earnings to be constant is a notational and mathematical convenience, but it will not affect the
results of the model.
   7 Conversely, if all college investment risk is idiosyncratic, investing in human capital is a way to diversify one’s

portfolio.
   8 In actuality, this choice is not binary, and over the time period of my analysis, time to degree has increased and

college completion rates have declined (Bound, Lovenheim and Turner, 2007). However, because this assumption
simplifies the model and conforms with my data, I maintain it throughout. With endogenous degree time and college
non-completion, the basic intuition and results of the model remain, but the solution becomes significantly more
complex.




                                                           6
plus forgone wages:

                                                         T                        T
                              C ≡                            Y0 e−rm t dt +           pe−rm t dt
                                                     0                        0
                                        1
                                     =    (Y0 + p)(1 − e−rm T )                                     (1)
                                       rm

   In this setup, the net pecuniary returns to investing in college are

              ∞                          T                           1 i −rm T    p
                  Yci e−rm t dt −            pe−rm t dt =              Yc e    −    (1 − e−rm T )   (2)
           T                         0                              rm           rm

   and the pecuniary returns to not investing in college are:

                                                 ∞                        1
                                                     Y0 e−rm t dt =         Y0 .                    (3)
                                             0                           rm

The optimal investment decision will be to undertake the investment if the net returns
to doing so are greater than the opportunity cost of the funds needed to finance the
investment:


                                    Y0 ≤ Yci e−rm T − p(1 − e−rm T )                                (4)


Re-arranging and taking logs, equation (4) becomes:

                                                               i
                                                            +p
                                                     ln( Yc +p )
                                                         Y0
                                                                     ≥ rm                           (5)
                                                              T

Equation (5) is similar to the well-known formula for optimal education investment,
although it shows the optimality condition for the binary decision rather than the
continuous investment decision. The formula implies an individual will find it optimal
to invest in college when the private rate of return to the investment is greater than
the market return from investing in an asset of identical risk, rm . This is the return
the individual would get if he did not undertake the college investment and instead
invested the savings from earnings. Crucially, the optimal investment decision is not



                                                                    7
a function of family resources.
    Borrowing constraints in this model come about because, while individuals can
invest at rate rm , they cannot necessarily borrow at rate rm . Consider the supply
of funds (SOF) to individual i who wishes to invest in college. The SOF comes
from the N assets available to each family. One such asset may be federal loans.
Another such asset is income and private savings, which can be borrowed at rate
rm . Housing equity is another source of funds, and the interest rate for borrowing
against this equity may be higher or lower than rm (taking into account the fact that
mortgage interest is tax-deductible). This treatment of the supply of funds assumes
no intra-household bargaining.9
    Given the assumption of full intergenerational asset transmission, at what interest
rate can each individual fund her education? If individuals could borrow against
                                                                                                        Y i +p
                                                                                                    ln( Yc +p )
future income (i.e., against human capital), then all students for whom                                 T
                                                                                                         0
                                                                                                                  ≥ rm
would be able to borrow C at interest rate rm as this is the rate if return on wage
income and savings. However, there is a built in borrowing constraint in higher
education because individuals cannot borrow against their future wages, so only
liquid assets at the time of college investment can be used to fund the investment.
In what cases will this constraint bind? First, note that because of the ordering of
the interest rates, each family will fully exhaust each asset before moving on to the
next, from 1, . . . , N . For example, the family will likely first exhaust government-
subsidized loans, such as Stafford loans, then use collateralized loans, such as home
equity loans and home equity lines of credit, and then rely on private savings as well
as uncollateralized loans such as Federal PLUS loans and credit card debt.
   9 For simplicity, I assume parents will fund their child’s education if it is optimal for the child to undertake the

investment. There is a recent literature on intra-household transfers and their relationship to education investment
(Brown, Scholz and Seshadri, 2007 and Brown, Mazzocco, Scholz and Seshadri, 2007) that presents evidence this
assumption is not valid for all individuals. To the extent children do not have access to the panoply of their parents’
resources, my empirical estimates will be biased towards zero.




                                                          8
   The average interest rate for each consumer will be:

                                           N
                                                 di
                                                  j
                                  ¯
                                  ri =           N          rj ,                                      (6)
                                           j=1   p=1   di
                                                        p


where


                    
                    
                                                 j
                    
                                     Wj    :     s=1   Ws < C
                    
                    
                                j−1               j−1                       j
               dj =  C −       s=1   Ws    :     s=1   Ws ≥ C >            s=1   Ws                  (7)
                    
                    
                    
                    
                                      0 : C<               j−1
                                                                   Ws
                                                            s=1


is the amount borrowed from source j.
   Thus, while the individual’s optimal decision depends on the pecuniary returns to
college investment and the market interest rate rm , her ability to act on this optimal
                          ¯                            ¯
decision is a function of ri . Note that regardless of ri , no individual will attend
college if equation (5) does not hold because the individual returns are lower than
one could get in the marketplace. Lowering the cost of funds does not change the
                                                                           Y i +p
                                                                        ln( Yc +p )
fact that the opportunity cost of funding college when                      T
                                                                             0
                                                                                      < rm is higher than
the pecuniary returns from the investment.
   In the case where equation (5) holds with equality, there are three distinct possi-
bilities:

     ¯
  1. ri < rm : In this case, the individual’s average cost of funds is lower than the
      market rate of return. The individuals will undertake the investment, and the
      relatively low cost of funds simply increases the net rate of return.

  2. ri = rm : If the market rate of interest equals the average cost of funds, then the
     ¯
      cost of funds does not cause a distortion in college investment nor does it alter
      the net rate of return.

     ¯
  3. ri > rm : In this case, the cost of funds is higher than the market interest rate,
      so the investment is individually optimal but students lack access to sufficiently

                                                 9
     inexpensive credit to fund the investment. I will refer to this case as “binding
     credit constraints,” because the prohibition against borrowing against future
     income is binding.

   An implication of this definition of credit constraints is that if the credit con-
straints bind, college investment will be sensitive to changes in asset values used
to fund college at the time of the investment decision. For example, take a simple
case in which a family has access only to government-subsidized student loans (r1 ),
housing equity (r2 ), and private loans (r3 ); it has no private cash savings. Further,
                                                                                     Y i +p
                                                                               ln( Yc +p )
assume the student finds college investment just optimal (rm =                        T
                                                                                      0
                                                                                              ).
   Figure 1 depicts this situation. In the figure, the saw-toothed curve is the marginal
cost of funds (MCOF), which increases discontinuously after each source of funds is
exhausted. The curved line shows the average cost of funds (ACOF) given the MCOF
curve. The average interest rate at which the individual can fund C is where the
ACOF curve (i.e., the supply curve of funds) hits the vertical line at C (i.e., the
demand curve for funds). The solid lines show the original position in which the
                                                  Y i +p                 Y i +p
                                               ln( Yc +p )            ln( Yc +p )
student is credit constrained: though rm =         T
                                                    0
                                                               ¯
                                                             , ri >       T
                                                                           0
                                                                                    , so the constraint
on borrowing against future human capital binds and the student does not invest in
college.
   Now, consider what happens when the value of the family’s house increases. This
increases the amount of equity against which the family can borrow; it has received
an increase in liquidity due to the increase in its home price. This scenario is shown
with the dotted lines in Figure 1: an increase in the value of the house extends the
amount of borrowing one can do at rate r2 , which lowers the ACOF curve such that
investing in college is now affordable for the individual. A reduction in the interest
rate on home equity would cause a similar effect, shifting the ACOF curve downward.
Finally, it is important to note Figure 1 assumes housing wealth is liquid. If housing
wealth were not liquid, it means either that the implicit interest rate on home equity
is very high or that the effective amount of home equity (W2 in this example) is zero.

                                          10
    Figure 1 illustrates that there are likely to be many inframarginal individuals.
For example, if a potential student has a return to education investment above r3
in Figure 1 (likely due to high ability), his investment decision will be unaltered
by changes in home equity. Furthermore, as previously discussed, any individual
for whom equation (5) does not hold also will be inframarginal because he will not
make the college investment. If there are marginal students affected by housing
price changes, however, Figure 1 shows how one can use housing price and wealth
variation to identify credit constraints. Only those who are credit constrained as
defined above will respond to housing price changes. If college investment behavior
responds to housing prices, and if it does so increasingly with the liquidity of housing
wealth, this will be evidence of credit constraints in college enrollment.10
    One important aspect of the definition of higher education credit constraints I use
in this analysis is that it is not dependent on wealth variation being predictable or
unpredictable. As long as education is purely an investment good, there should be no
wealth effect. Enrollment responses to any wealth variation will indicate the existence
of credit constraints. The intuition for this result is embodied in equation (5): the
optimal attendance decision is a function only of the internal rate of return and the
market interest rate, not family resources. Any dependance of attendance on family
resources therefore identifies higher education credit constraints. This definition is
in stark contrast to the definition of credit constraints used in the consumption
literature to test the permanent income hypothesis (Zeldes, 1989; Runkle, 1991;
and Jappelli, Pischke, and Souleles, 1998). For consumption goods, perfect capital
markets imply no behavioral response to expected wealth or income changes, whereas
for education, perfect capital markets imply no behavioral responses to any income
or wealth changes. The remainder of this paper seeks to identify the relationship
  10 It also is possible that increasing housing wealth causes higher college enrollment due to a wealth effect; if there

is consumption value to college, increasing wealth will increase attendance. There is considerable debate within
economics about whether housing wealth constitutes real wealth. It is likely housing wealth increases serve mostly
to increase household access to liquidity rather than represent real wealth because, if the home is sold, the family
must buy or rent a new home. The only way to realize housing wealth increases is to move outside one’s housing
market, but it is unlikely parents move in this manner in order to pay for their children’s postsecondary education.
I examine the evidence on whether I am identifying a wealth effect rather than a liquidity effect in Section 6.4.



                                                          11
between housing wealth and college enrollment empirically using this implication of
perfect capital markets for education.



3    Data

The individual-level data in this analysis come from the Panel Study of Income
Dynamics (PSID). The PSID began following a nationally-representative sample of
households in 1968 and has followed it and its descendants continuously since that
time. These data are particularly suited to address the central research questions
set forth in this paper because they contain information on educational attainment,
self-reported housing prices, housing equity, and a rich set of family background
characteristics. Because it is such a long panel, it is possible to link college students
with their parents to measure parental housing wealth and income.
    While there are PSID surveys available continuously between 1968 and 1997, after
1997 they were conducted every other year. To make a data set consistent with the
PSID sample timing, I construct a repeated cross section of 18-19 year olds from
each PSID survey in every second year beginning in 1980. This methodology also
has the advantage of maximizing the number of observations within each year, which
will help identify the year fixed effects in the regressions. The years included in the
analysis are 1980, 1982, 1984, 1986, 1988, 1990, 1992, 1994, 1996, 1999, 2001, 2003,
and 2005. The panel ends in 2005 as this is the most recent PSID survey year. Thus,
my analysis will cover the housing boom of the late 1990s through 2006 but will not
cover the recent slowdown in the housing market.
    On average, there are 560 observations in each year for a total of 7,276 observa-
tions. I include only 18 and 19 year olds in the sample, which ensures I do not double
count the same individual in subsequent years. Furthermore, respondents who are
18 or 19 years old are the ones making the college attendance decision and for whom
this decision is most likely to be dependent on their parents’ resources.



                                           12
    The PSID does not directly ask for college enrollment but contains information
on years of school completed. I measure enrollment as having completed more than
12 years of schooling. Note this definition of enrollment is somewhat different from
most of the college enrollment literature as I do not count students who attend college
but drop out before completing their first year. Because of the long nature of the
panel, I can determine whether a student completed a thirteenth year of education
in subsequent surveys follow-ups. If a student completed more than twelve years of
education within two years of the survey, I classify him as enrolled.11
    I use 5 different measures of housing wealth from the PSID. The first is contem-
poraneous self-reported housing prices. Self-reported housing values may be prob-
lematic because of incomplete information on the part of the owner or because of
measurement error. To check whether individuals systematically misrepresent their
home value, I construct a median and mean housing price index from the PSID self-
reported housing values data. Note these indices are for the full PSID sample, not
just for the sample of households with 18 or 19 year olds. To construct the indices,
I calculate the mean and median housing price from the PSID in each year, then
divide all values by the value in 1980. Figure 2 compares these indices to the Con-
ventional Mortgage Housing Price Index (CMHPI),12 which I have rescaled such that
1980 equals 100. It is clear from the figure that the the aggregate median and mean
reported housing values in the PSID track the national index quite closely. That the
mean PSID index diverges slightly in more recent years is due to the fact that new
home sales and jumbo loans are not included in the CMHPI, but they are included
in the PSID measures. Figure 2 suggests, if there is measurement error in the PSID
housing prices, it does not show up in the aggregate trends.
  11 For the 2005 survey, this methodology is not possible. Thus, the enrollment levels in the 2005 PSID are slightly

lower than in the previous survey waves. I include year fixed effects in my regressions in part to deal with this
problem.
  12 The CMHPI is an index of all mortgages that were purchased or securitized by Fannie Mae or Freddie Mac

over the course of a year. These mortgages are conventional in the sense that they are not insured by the federal
government and are for single family homes only. The CMHPI also does not include “jumbo loans,” which in 2007
was any loan above $417,000. While the index does include appraisals from refinances, it does not include price
data from new homes. Instead, it is based solely on repeat transactions. The Office of Federal Housing Enterprise
Oversight (OFHEO) puts out a similar housing price index that includes mortgages from other lenders. The two
indices are quite similar and estimates using the OFHEO index are identical to those using the CMHPI index.


                                                         13
  In addition to contemporaneous housing prices, I construct contemporaneous
housing wealth measures. The ability to measure housing wealth rather than just
housing prices is a major advantage of the PSID, because if a homeowner has no
equity built up in her home, the value of the house represents illiquid wealth that
is irrelevant for the purposes of financing college education. The PSID asks about
the amount of principal remaining on the loan. The value of the home minus the
remaining mortgage principal is the accumulated home equity. This amount can be
negative, particularly for newer home loans where the borrower has put 0% down
or if the value of the home drops below the remaining mortgage principal. I also
use the change in home equity and home values over the 4-year period prior to the
survey date. I construct these variables by calculating home equity and home prices
4 years prior to a given survey year, then taking the difference between the current
and lagged values.
  The strengths and weaknesses of these measures and the implicit identification
assumptions underlying their use is discussed in detail in Section 5. However, I
construct a fifth housing wealth measure motivated by the fact that the growth in
housing wealth may be endogenous to college attendance. If parents tap their home
equity to invest in college, home equity growth and college attendance will be nega-
tively correlated in the data. I solve this problem by estimating the counterfactual
growth in home equity. This counterfactual growth is the amount of growth expected
if the household had continued paying its mortgage over the 4-year period without
tapping into its equity. This variable is intended to measure the amount of equity
growth the household would expect to have over this period.
  The central difficulty in constructing the counterfactual equity level is estimating
the equity accumulation over the 4-year period. Given full information about the
loan, this calculation is simple. However, I do not observe the interest rate or the
term of the loan, but I do observe the yearly amount of mortgage payment and the
remaining principal on the loan. I use the ratio of the mortgage payment to the


                                         14
remaining principal to estimate the interest rate, which I then use to calculate the
equity accumulation the household would expect over the previous 4 years.
   Estimation of the interest rate proceeds as follows: first, I assume all loans have
a 30-year term. I use the national average mortgage interest rate on single family
homes reported by Freddie Mac and assign an interest rate to each loan age in each
survey year. For example, in the 2001 survey, a loan that was 10-years old, thus
originating in 1991, was assigned an interest rate of 9.25%. In 2003, a loan that was
12-years old would have the same interest rate, whereas a loan that was 10-years old
would have an interest rate of 7.31%, which was the interest rate in 1993.
   For each year and mortgage age from 0 to 30, I calculate the ratio of the monthly
mortgage to remaining principal implied by the interest rates I assigned to each
year-mortgage age combination. I then calculate this ratio for each respondent. The
interest rate and mortgage age combination that minimizes the squared difference
between the two ratios identifies the imputed interest rate. Using this interest rate,
I calculate the equity each household with a given housing wealth in period t-4
would accumulate over the next 4 years. The central difference across households is
generated by where each household is in its loan repayment schedule. The current
value of the home minus this counterfactual equity amount is the counterfactual home
equity in year t. Subtracting actual home equity in t-4 from counterfactual home
equity in year t yields the amount of home equity each household would accumulate
over the 4-year period if it did not tap any of the equity in its home.
   The PSID also contains detailed demographic information about each respondent
and household. I construct measures of the household head’s education level, age,
marital status, and sex, the number of other dependents under 18 living in the house-
hold, the respondent’s race and gender, and total family income from all sources.
These variables are taken directly from the PSID and are measured as of the current
year of the survey. Notably, I have no information on financial aid received by each
household. However, to the extent financial aid policies alleviate credit constraints,


                                          15
this ommission will bias my housing wealth estimates towards zero.13



4     Housing Wealth, Income and College Enrollment

Before undertaking the empirical analysis of the importance of housing wealth in
college enrollment, it is instructive first to examine simple tabulations of the preva-
lence of homeownership among college-going families. Figure 3 presents the rate of
homeownership over time for the population of families with 18-19 year olds in the
October CPS. Homeownership rates vary over time between 85% and 90% and vary
little between 2-year and 4-year students.
    Figure 4 shows the rate of college attendance for children of homeowners and
renters separately over this same time period. As expected, college attendance is
higher among the homeowner population, and this gap is larger within 4-year schools
than within 2-year schools. Furthermore, since 1977, college enrollment has increased
by 11.7 percentage points (or 19.7 percent) among homeowner families but has de-
creased slightly for renters. While much of these gaps can be explained by differences
in family resources and student ability that are correlated with homeownership, Fig-
ures 3 and 4 indicate that housing wealth may be an important component of the
college attendance decision.
    The previous literature examining the income gradient in college attendance and
academic attainment largely has ignored housing wealth (Ellwood and Kane, 2000;
Shea, 2001; Carneiro and Heckman, 2002; Cameron and Heckman, 1998 and 2001;
Plug and Vijverberg, 2005; Brown, Scholz and Seshadri, 2007; Belley and Lochner,
2007). Omitting household wealth ignores a potentially important aspect of house-
hold resources that can be used to finance college. As housing prices grew during the
millennial housing boom, housing wealth likely became a more important component
  13 Omitting financial aid will cause a negative bias in my estimates as long as college attendance and financial aid

are positively correlated, as argued by Dynarki (2002), and financial aid and housing wealth are negatively correlated,
conditional on the other observables in the model. The latter correlation is likely to be negative because housing
wealth is “taxed” for the purposes of institutional aid, even though it is not for the purposes of federal aid after 1992.




                                                           16
of household assets. The large rise in housing wealth over my sample period is shown
in Figure 5, which presents real average home prices, home equity and counterfactual
home equity in each year of the analysis.14 Several relevant trends become apparent
in Figure 5. First, real home prices increased by 77 percent across the entire sample
between 1992 and 2005, with the most significant growth between 2001 and 2005.
Second, housing equity grew less rapidly than home prices, suggesting not all of this
growth was capitalized into wealth. Finally, while housing equity and counterfactual
housing equity track each other closely through the 1990s, in the 2000s, counter-
factual housing equity rose faster than housing equity. This differential growth is
consistent with households tapping their equity in the time period surrounding when
their children become of college age in the 2000s.
   Depending on the joint distribution of housing wealth and income, ignoring hous-
ing wealth can cause the income coefficients in college attendance regressions to be
biased in either direction. Table 1 shows the joint distribution of housing wealth
quartiles and total family income quartiles. The quartiles are constructed for all
homeowners in the PSID sample discussed in the previous section. Panel A shows
the joint distribution with respect to housing prices, and Panel B shows the joint
distribution with respect to home equity levels. If housing wealth and family income
were perfectly correlated, the diagonals would be equal to the average percent at the
end of each column. However, in Panel A, only 42.8% of households are in the same
income quartile as their home value quartile. Another 15.8% have a home value in a
higher quartile than their income, and 41.43% have a lower home price quartile than
income quartile. Similar patterns are evident for home equity levels. There are signif-
icant numbers of high-income, low-housing equity households as well as low-income,
high-housing equity households, suggesting ignoring housing wealth may cause an
understatement of the importance of family resources in college attendance.
   Table 2 presents similar joint distributions to Table 1, but instead of using hous-
 14 All   references to “real” prices refer to prices that have been adjusted to 2007 dollars using the CPI-U.




                                                           17
ing levels uses quartiles of the change in housing measures. Panels A-C show the
change in housing values, housing equity, and counterfactual housing equity levels,
respectively, over the four-year period prior to each survey year included in the PSID
sample. The distributions in Table 2 are even less concentrated on the diagonal than
those in Table 1. For example, in Panel C, only 36.1% of households that own a home
are in the same family income quartile and the same quartile of the change in their
counterfactual home equity. Over 23.2% are in the upper diagonal, and 40.7% are
in the lower diagonal. Housing wealth and housing wealth changes are not perfectly
correlated with income, and income is even less correlated with changes in housing
wealth than with the level of housing wealth. In other words, changes in housing
wealth have occurred relatively evenly across the income distribution, suggesting the
importance of this form of wealth, particularly for low-income households wishing to
finance a college education.



5   Empirical Models

The goal of the empirical analysis is to identify the effect of access to housing wealth
on the propensity to enroll in college. I estimate the following models using the PSID
data described in Section 3 that allows for housing wealth to have a different effect
on college enrollment across decades:


 Enrollit = β0 + β1 Own + I(1980s)[ψ1 Hit + φ1 Yit + α1 Mit ] + I(1990s)[ψ2 Hit (8)

               +φ2 Yit + α2 Mit ] + I(2000s)[ψ3 Hit + φ3 Yit + α3 Mit ] + γXit + δt +     it




    Enrollit = β0 + β1 Own + I(1980s)[ψ1 (Hit − Hi,t−4 ) + φ1 Yit + α1 Mit ]              (9)

                  +I(1990s)[ψ2 (Hit − Hi,t−4 ) + φ2 Yit + α2 Mit ]

                  +I(2000s)[ψ3 (Hit − Hi,t−4 ) + φ3 Yit + α3 Mit ] + γXit + δt +   it ,




                                          18
where Enroll is a dummy variable equal to 1 if the student enrolls in college and
Own is a dummy variable equal to 1 if the household owns its own home, H refers
to housing prices, and Y is real total family income from all sources. Because
each household’s mortgage is a fixed consumption commitment (Chetty and Szeidl,
2007), the mortgage payments reduce disposable income that can be used for college
tuition.15 I control for yearly mortgage payments (M ) in order to account for this
effect in equations (8) and (9), setting M equal to zero for renters.16 Equations
(8) and (9) also include a vector of demographic and household characteristics (X )
discussed in Section 3 as well as year fixed effects, denoted by δt .
    Equations (8) and (9) differ in how housing prices and wealth (H ) are measured.
In equation (8), H refers to contemporaneous home price or housing equity of the
household. Using contemporaneous measures is problematic because housing wealth
is likely to be correlated with unobserved ability of the prospective students. This
correlation will cause an upward bias in β2 , assuming that unmeasured ability and
housing wealth are positively correlated conditional on the other observables in equa-
tions (8) and (9) and that higher ability students are more likely to attend college.
These assumptions are consistent with the existing evidence on wealth, income, abil-
ity and college attendance (Carneiro and Heckman, 2002; Belley and Lochner, 2007).
    One potential solution to the omitted variables problem caused by unobserved
ability is given in Equation (9), which includes the four-year growth in housing
prices and equity. The implicit “experiment” underlying this regression is some
people own homes in high-growth areas and some own homes in low-growth areas.
For example, between 1980 and 2005, San Francisco witnessed a sixfold increase in
housing prices, while Oklahoma City prices barely doubled and were lower than San
Francisco prices in absolute terms. While locational decisions of households are non-
  15 Other expenditures, such as food, also represent consumption commitments because one must eat to live. How-

ever, for most homeowner families with college-age kids, conditional on not selling one’s home, there is little control
over the yearly mortgage payment levels. The same cannot be said for food expenditures or most other consumption
goods, which is why I do not control for these expenditures in my empirical models.
  16 The PSID conducted wealth supplements in 1984, 1989, 1994, 1999, 2001, 2003 and 2005 that contain information

on non-housing wealth. I ran equations (8) and (9) including non-housing wealth for these years and results were
unchanged. These results are available from the author upon request.



                                                          19
random, the central identifying assumption in equation (9) is that the rich set of
background characteristics of the household is sufficient to control for such selection.
In other words, conditional on these observable characteristics, the growth in house-
hold housing wealth is exogenous. That supply restrictions are an important driver
of housing price growth further supports this argument (Glaeser, Gyourko and Saks,
2005; Gyourko, Mayer and Sinai, 2006). Additionally, as Table 2 illustrates, there
was a large amount of variation in the magnitude of housing price increases across
the income distribution, which is at least suggestive that housing price increases are
more random with respect to ability than contemporaneous levels.
    Equation (9) includes both the growth in housing prices and housing equity. As
previously argued, the focus on equity is desirable because a household with a high-
priced home but no equity has little liquid housing wealth. However, the growth in
housing wealth is likely endogenous with respect to the college enrollment decision.
If parents pay for college by tapping their home equity, then households with enrolled
students can have lower equity growth than other households.17 I use the counter-
factual growth in home equity to solve this problem. The counterfactual growth in
home equity can be expressed as follows:


                                        ˆ
                                 (Pt − M P t ) − (Pt−4 − M Pt−4 ),                                            (10)


                                                  ˆ
where MP is the remaining mortgage principal and M P is the counterfactual re-
maining mortgage principal in time t expected if the household did not tap any
housing equity between t-4 and t. This measure of the growth in housing wealth is
independent of any endogenous behavior the household engages in over the 4-year
period. Comparing results across specifications employing actual and counterfactual
home equity growth yields insight into the degree to which the observed growth is
endogenous.
  17 Conversely, if a household experiences a large, unexpected, temporary positive income shock, it may pay off some

of its mortgage debt, thereby increasing home equity, and be more likely to be able to finance a college education.




                                                        20
    Equations (8) and (9) allow the effects of housing wealth and income to be dif-
ferent across decades. An advantage of comparing effects over time is it allows one
to assess the severity of any omitted ability bias. Belley and Lochner (2007) present
evidence from the NLSY79 and NLSY97 that the ability gradient in enrollment, con-
ditional on family income and demographic characteristics, has changed little over
time. Estimates from that paper are reproduced in Table 3, which also includes a
similar comparison across cohorts from the NLS72 and NELS:88 surveys.18 Compar-
isons in Belley and Lochner (2007) use AFQT scores as the ability measure, while
the results from comparisons of NLS72 and NELS:88 respondents use the NCES-
administered math and reading scores as the ability measures. Both sources show,
insofar as ability and income are correlated, this correlation has remained constant
over time. While neither data source allows an analysis of household wealth, these
results are suggestive that the relationship between ability and housing wealth may
have remained constant over time as well.
    There was also a stark change in financial markets that occurred over this time
period; in the late 1990s, housing wealth became more liquid. While there is much
debate over why it became easier to extract one’s home equity, that home equity
loans, home equity lines of credit, and cash out refinances grew is not in question.
Figure 6 shows the total amount of extracted home equity (Greenspan and Kennedy,
2005) as a ratio of total personal income in the United States between 1990 and 2004.
The figure also includes the ratio of median home price to average per capita income.
The increases in home equity extraction as a proportion of real income are striking;
in 1990 home equity extraction was twice real personal income, while in 2004 it was
over 9 times real personal income. Home price increases, while substantial over this
period, are much smaller in comparison.
    Because illiquid wealth should have little effect on college enrollment if the only
  18 See Bound, Lovenheim, and Turner (2007) for a complete description of these data. While they do not show the

tabulations presented in Table 3, the data used to generate the NLS72 and NELS:88 results are identical to the data
used in their analysis.




                                                        21
mechanism linking the two is financial, comparing the relationship between housing
equity and college enrollment across periods of differential home equity liquidity is
instructive. If housing wealth has a large effect on college enrollment in the 1980s,
when housing wealth was relatively illiquid, it may point to a potential omitted
variables bias in my estimates. On the other hand, if the estimated effects match the
trends in equity extraction, it will suggest effects of omitted ability may be small.
    Examining changes in the effect of housing wealth on college enrollment across
decades also assumes households are not increasing the amount of housing wealth
they hold in response to expected increases in future housing liquidity. Note that
an unconstrained homeowner never has an incentive to increase his home equity to
borrow from this equity in the future. The intuition for this result is that, conditional
on owning a home, the marginal dollar allocated to the home will be more expensive
to borrow against in the future relative to liquid wealth due to the forgone interest on
this dollar. Any changes in college investment behavior due to changes in the liquidity
of housing wealth can, therefore, be attributed to capital market imperfections.19
    A final reason for allowing estimated housing wealth effects to differ across decades
is the increasing cost of attending college. Panel A of Figure 7 presents average
tuition for public 4-year schools, private 4-year schools and 2-year schools as a percent
of average real per-capita income between 1977 and 2005. In 1980, average tuition
at four-year public schools was 7 percent of per-capita income20 and was 31 percent
of per-capita income at four-year private schools. By 2005, tuition and fees alone
were more than 20 percent of average per-capita income at four-year public schools
and rose to almost 80 percent at four-year private schools. Even at two-year colleges,
tuition costs as a percent of annual income more than doubled over this time period,
  19 Equations (8) and (9) also assume households are not endogenously placing more of their wealth in their home

to take advantage of expected housing price growth or because after 1992, housing wealth was exempted from federal
financial aid calculations. Reyes (2008) shows that the federal financial aid tax has no effect on the amount of
housing equity held by households, which makes sense considering that housing wealth is still taxed for purposes of
institutional aid. Furthermore, because one does not need to increase one’s equity position in a home to obtain the
equity benefits of a given price increase, it is unlikely households respond to future expected housing price growth
by putting more of their liquid wealth into housing, conditional on owning a home.
  20 I find changes in 4-year public in-state tuition levels are uncorrelated with changes in state average home prices,

conditional on state fixed effects and state-specific linear year trends over the sample period.



                                                          22
from 3.6 to 8.6. Panel A shows statutory tuition levels, which overstate the cost
of attending college due to financial aid. Panel B of Figure 7 shows that the net
cumulative cost of attending a 4-year public school has increased markedly since
2000. The figure shows the total average cost (including room and board) of 4-
year attendance and the total amount of the maximum Pell Grant and Stafford
loan amounts, which are the predominant forms of Federal financial aid. The net
cumulative cost is the total cost net of average grant aid and tax deductions summed
over the four years starting in the year indicated on the x-axis. Panel B illustrates
that after 1990, federally subsidized loans and grants were insufficient to cover the
cost of attending a four-year public university, and as federal aid has decreased in
real terms, the net cost of attendance increased by almost 30 percent between 2000
and 2005. Furthermore, because many middle class families are ineligible for Pell
Grants and subsidized Stafford loans,21 and because these families are likely to own
a home, it is reasonable to expect that their reliance on housing wealth to finance
college has increased over this time period.
    While there are no published aggregate trends of home equity extraction to pay for
college, the increased reliance of many homeowners on their home equity is consistent
with the change in borrowing patterns among college attendees. Between 1996 and
2005, real subsidized Stafford loans increased by 17 percent, but real unsubsidized
Stafford loans increased by 94 percent and PLUS loans increased by 177 percent
(College Board, 2007). Figure 8 shows the historical interest rates associated with
home equity lines of credit (HELOC), Stafford loans, and PLUS loans. Since 1985,
when interest rates for federal financial aid fell below the statutory caps, these rates
have move similarly. Most notably, federally-backed PLUS loans, which are loans
designed to cover the remaining cost of college after all financial aid is accounted
for, and HELOC loans have almost identical interest rates. Due to the similarity
  21 There are two types of Stafford loans: subsidized and unsubsidized. While both types have the same interest

rate, repayment of subsidized loans can be deferred until after college without accumulating interest. Both types of
Stafford loans are backed by the federal government.




                                                        23
of pricing between PLUS and HELOC loans, it is probable that home equity loans
among college-going families have grown similarly.
    Another potential identification concern with equations (8) and (9) is that housing
price growth may be correlated with local labor market conditions, in particular
demand for high-skilled labor. Any positive correlation between high-skilled labor
demand and housing prices will serve to bias upward estimates of the effect of housing
wealth on enrollment. If my estimates are contaminated by this correlation, the
contamination should occur similarly across wealthier and poorer households. I use
this fact to conduct robustness checks of my estimates by splitting my sample by
income and housing wealth into those who are more and less likely to be credit
constrained. If increasing housing equity serves to relax credit constraints, only the
constrained group should be affected, whereas if high-skilled labor demand raises
both housing wealth and enrollment, both groups should be affected.


6     Results

6.1     Parameter Estimates

Table 4 presents logit estimates of equations (8) and (9) using the PSID data de-
scribed in Section 3.22 The columns differ in the housing measure included in each
regression. Column (i) contains results using contemporaneous home prices and col-
umn (ii) uses contemporaneous home equity. Columns (iii) and (iv) show results
from regressions in which the 4-year change in home prices and home equity are
used, respectively. Column (v) contains estimates from the counterfactual housing
equity change specification.
    Table 4 shows the effect of housing wealth on college enrollment is isolated to
the 2000s. Across columns, there is no evidence that housing prices or home equity
were correlated with college enrollment in the 1980s or 1990s; all point estimates are
  22 Results in Table 4 include controls for household head’s education level, age, sex and marital status, respondent’s

age, sex and race, and the number of other dependents in the household, but I have not reported demographic
coefficients for brevity. All regressions are weighted by the family weights in the PSID.


                                                          24
close to zero and none are significant at even the 10 percent level. In the 2000s, a
$10,000 increase in housing prices increases the log odds ratio of college enrollment
by 0.010, which is significant at the 10 percent level. However, a $10,000 increase in
housing prices in the 4 years prior to a child turning 18 or 19 increases the log odds
ratio of college attendance by 0.019. This estimate translates into an elasticity of
0.011 and is significant at the 5 percent level. While this elasticity is small, the large
increases in home prices since the 1980s imply even a small effect will be economically
meaningful. For example, as Figure 2 illustrates, home prices nationally increased by
over 360 percent between 1980 and 2005. My results suggest these increases would
increase college enrollment by nearly 2 percent.23
    While the relationship between observed home equity or the change in home equity
and college enrollment is not strong in the results presented in Table 4, focusing on
the change in counterfactual home equity suggests housing wealth is important in the
college attendance decision. The coefficient on the counterfactual growth in housing
equity is 0.028, which translates into an elasticity of 0.015 at the mean of all variables.
The fact that counterfactual home equity increases are more strongly associated with
college attendance than home equity levels or changes is consistent with homeowners
extracting their home equity to pay for their children’s college because the change
in home equity for those who enroll is smaller than for those who do not enroll.
    The pattern of effects shown in Table 4 also argue against omitted ability bias: if
the growth in housing wealth were just proxying for unmeasured ability, I should find
effects in the 1980s as well as in later periods. The coefficients are consistent with
there being no effect in the 1980s, and the 1980s coefficient and 2000s coefficient are
statistically different at the 5 percent level for the growth in counterfactual housing
equity. Rather than measuring ability, the estimates track closely the trends in
equity extraction over the past several decades as shown in Figure 6. Thus, the
  23 These results are robust to controlling for state fixed effect. However, when I include state fixed effects, the

standard errors become noticeably larger because of small within-state sample sizes in many states. For this reason,
I exclude state fixed effects from my main specifications. These results are available from the author upon request.




                                                        25
central driving forces behind my results are the growth in home prices combined
with the increasing liquidity of housing wealth and the growing real cost of college
(see Figure 7).
   The estimates in column (v) of Table 4 have important policy implications due
to the recent housing bust in some areas of the country combined with the “credit
crunch” that has limited households’ access to their home equity. One way to assess
the economic significance of these results is to simulate what college enrollment would
be if housing prices fell by 18.2 percent for all households in the 4 years before it has a
child of college age. I chose 18.2 percent because this is the national percent decline
in home prices according to the Case-Shiller home price index since the housing
market peaked in June 2006. If all home prices declined by 18.2 percent, this would
cause a 4.0 percentage point decline in college enrollment due to the fact that an 18.2
percent decline in home prices represents a significant reduction in home equity for
most households. This exercise likely understates the effect of the current housing
market decline on college enrollment because it assumes the liquidity of housing
wealth has not changed. Given that the recent housing market slowdown has been
accompanied by a reduction in consumer access to home equity loans and home
equity lines of credit, the effect on college enrollment could be much larger than I
estimate.
   I also can use the estimates in Table 4 to construct an upper bound on the number
of households for which housing wealth relaxes credit constraints in paying for col-
lege. Using the estimates from column (v), I simulate what college enrollment would
be if all housing wealth became illiquid. I perform this calculation by assuming each
homeowner looses all of the equity in his home in the four years prior to his child
becoming of college age. If all housing equity were to become illiquid, I estimate
college enrollment would decline by 6.8 percentage points. This admittedly extreme
simulation represents an upper bound on the percent of households that use housing
wealth to relax credit constraints because it assumes households will not endoge-


                                            26
nously shift their resources to compensate for the change in their housing wealth.
This back-of-the-envelope calculation suggests housing wealth serves to relax credit
constraints for a small but non-trivial portion of the college-going population.


6.2   Sample Splits

Aside from omitted student ability, a potential concern in my estimates is the source
of variation in home prices. As discussed in Section 5, if prices respond to local
labor market shocks, the coefficient on the change in housing wealth will be biased.
Table 5 tests for such effects by splitting the sample into those who are more and
less likely to be income constrained. The first two columns split the sample into
households with real income over and under $125,000, the second two columns split
the sample into those with counterfactual home equity above and below $125,000,
and Columns (v) and (vi) split the sample into households that have a ratio of real
income to counterfactual home equity over and under 1.5. The last split is similar
to those used by Zeldes (1989), Jappelli, Pischke and Souleles (1998), and Souleles
(2000).
   Results from these sample splits are shown in Table 5. The estimates are consis-
tent with the existence of credit constraints: the effect of housing wealth growth on
college enrollment is restricted to households with income under $125,000, counter-
factual housing equity under $125,000, and with income less than 1.5 times the level
of counterfactual housing equity. Furthermore, the estimates are only statistically
significant in the 2000s, which is consistent with the results in Table 5. The results
from the sample splits are consistent with the existence of credit constraints but not
consistent with the endogeneity of housing wealth growth due to local labor market
shocks.
   To examine further the source of variation in housing prices and housing wealth,
I use the restricted PSID that contains MSA identifiers to estimate effects separately
for MSAs that more likely experienced exogenous housing price increases. Similar


                                          27
to Gyourko, Mayer and Sinai (2007), I identify MSAs that experienced relatively
high housing growth and relatively low population growth. These MSAs more likely
experienced exogenous housing price growth because, as Gyourko, Mayer and Sinai
(2007) argue, the housing prices were coming through supply constraints rather than
demand shocks that could be correlated with college investment behavior.
    I combine population estimates from the U.S. Census Bureau and use MSA-level
CMHPI housing price indices from Freddie-Mac that contain prices from MSA-level
repeat home sales. Note that the housing price growth estimates using the CMHPI
are conservative because they exclude jumbo loans. I classify an MSA as low popula-
tion growth and high housing price growth if annualized housing price growth either
between 1990 and 2005 or between 2000 and 2005 was above the median for all MSAs
and if annualized population growth was below the median. This methodology leads
to 51 MSAs classified as high housing and low population growth using 1990-2005
growth rates and 48 MSAs so classified using 2000-2005 growth rates.24
    Columns (i)-(iv) of Table 6 shows estimates of equation (9) by MSA growth type.25
For the low population, high housing price growth MSAs, results follow a similar
trend across decades, but the estimates in the 2000s are over double those shown in
Table 4 for the full sample. The effect of housing equity growth on the log odds of
college enrollment is 0.053 using the 1990-2005 annualized growth rates and is 0.043
using the 2000-2005 annualized growth rates. Of equal importance, columns (ii) and
(iv) show that results from all other MSAs are smaller and not statistically significant.
The results in Table 6 illustrate that all of the estimated effect of housing wealth
growth on college enrollment in Table 4 is coming from respondents in MSAs that
most likely experienced exogenous housing price increases due to supply constraints.
    The results presented thus far use housing wealth variation at the individual
  24 See Appendix Table A-1 for a list of low population growth, high housing price growth MSAs. As the table

shows, California cities make up a large number of these samples. All reported results in this paper are robust to
dropping respondents from California. The results are also robust to excluding Texas, which did not allow home
equity loans prior to 1997.
  25 Results in Table 6 exclude the 20 percent of the sample who do not live in an MSA. All results reported in Tables

4 and 5 for the full sample are similar in sign, magnitude and statistical significance when I restrict the sample to
those living in an MSA.



                                                         28
household level. However, as suggested by columns (i) through (iv) of Table 6,
much of the housing price variation is likely occurring at the MSA-level. As a final
robustness check, I use MSA-level housing price variation from the CMHPI to predict
housing wealth changes for each household if its home price varied perfectly with
average housing prices in its MSA. Household i’s home price in time t conditional on
its t − 4 housing price and the average MSA-level housing price growth is:

                                           cmhpijt
                           Pˆ = Pij,t−4 ∗
                            ijt                      .                            (11)
                                          cmhpij,t−4

The associated counterfactual growth in home equity is then:


                          ˆ     ˆ
                         (Pt − M P t ) − (Pt−4 − M Pt−4 ).                        (12)


   Note that equation (12) is a function solely of t − 4 home prices and remaining
mortgage principal.
   Column (v) of Table 6 show the results of estimating equation (9) using MSA-
level housing price growth. Similar to the results using individual-level variation,
column (v) of Table 6 shows that housing equity growth as predicted by MSA-level
housing price growth affects college enrollment only in the 2000s. The similarity
of these estimates with those in column (v) of Table 4 suggests MSA-level housing
price variation is driving identification of the relationship between housing wealth
and college enrollment. This result is reassuring because it indicates individual-level
measurement error in housing prices is not identifying my estimates. Taken together,
the results from Table 6 suggest the positive effects of housing wealth on college
enrollment are concentrated in the 48 to 51 MSAs that have experienced large price
increases relative to population growth and are coming through MSA-level housing
price increases.




                                          29
6.3     Evidence on Home Equity Extraction Among Households with College-
        Age Children

The implication of the results in Tables 4-6 is that some parents are tapping their
home equity to pay for college. Beginning in 1994, the PSID began asking whether
a family had a second mortgage in its wealth supplements, typically in the form of
a home equity loan or a home equity line of credit. Using the sample of all 18-22
year olds in the PSID linked to their families, I estimate the probability each family
has a second mortgage as a function of family characteristics, family finances, and
the 18-22 year old having ever enrolled in college.26 Results are reported in Table 7
and indicate that among all families with 18-22 year olds, those with a child who has
enrolled in college are more likely to have tapped their home equity.27 For example,
in Column (iii), which controls for real family income, non-housing wealth, and home
value, a household with a dependent enrolled in college has a 0.219 higher log odds
ratio of having a home equity loan or home equity line of credit. The Dependent in
College estimates are significant at the 5 percent level across all columns. While there
is no experiment to identify the causal parameters in Table 7, the results suggest that
families tap their home equity to pay for college, which is consistent with the results
reported in Tables 4-6.
    There is also some survey evidence on the use of home equity in financing college
costs. Next Step Magazine conducted a survey of parents with college-age children
and found nearly 25 percent reported they were planning to finance tuition using their
home equity. Further, about 3 percent of home equity loans in 2006, which translates
into about $7 billion, were taken out to finance higher education expenditures (Grant,
2007).
  26 I use the 18-22 year old sample because I am restricted to the PSID wealth supplements, which were conducted

in 1994, 1999, 2001, 2003, and 2005. I include these additional households to increase sample size, but the results
are qualitatively similar when I restrict the sample to 18-19 year olds. Furthermore, I am unable to test for different
effects across decades due to the limited number of survey years in the 1990s and the resultant small sample sizes.
  27 Note my definition of home equity extraction excludes cash-out refinancing. Hurst and Stafford (2004) examine

this form of equity extraction exclusively and find households use this financial tool to insulate themselves from
unemployment shocks. It is thus likely Table 7 understates the extent of equity extraction.




                                                         30
    That parents use their housing wealth to help pay for their children’s college
education has a significant amount of support, both in the PSID data and from other
sources. The results from this analysis suggest housing wealth is linked to college
enrollment through the cash-out mechanism, which implies college enrollment may be
increasingly tied to fluctuations in the housing market as families make savings and
investment decisions based on the expected availability of housing wealth. Shocks
to equity access, such as have occurred recently due to the slowdown in the housing
market, can have negative consequences for college enrollment that largely have been
ignored by policy-makers.


6.4     Credit Constraints or Wealth Effect?

As discussed in Section 2, one of the central assumptions underlying my identification
of credit constraints is that there is no wealth effect on college enrollment from
changes in housing wealth. If college attendance has consumption value, however,
increasing household wealth will increase consumption of college regardless of credit
constraints. This is a difficult assumption to test with my research design because
both wealth effects and higher education credit constraints will manifest themselves
similarly in the data.
    The results presented thus far are arguably more consistent with credit constraints
than wealth effects. If households were to increase college consumption due to hous-
ing wealth increases, one would expect to see this behavior in the 1980s and 1990s.
Similarly, one might expect both poorer and wealthier households to increase college
attendance if wealth effects were the dominant explanation for the positive housing
wealth, college enrollment relationship. The results from Tables 5 and 6 run counter
to such patterns, but depending on the shape of the Engel curve for college consump-
tion28 (absent credit constraints) and on how households responded to increases in
less liquid housing wealth in the 1980s and 1990s, these results do not rule out wealth
  28 Interestingly, Belley and Lochner (2007) find a positive effect of income on college attendance among households

in the top half of the wealth distribution.



                                                        31
effects.
   Another way to examine whether I am identifying wealth effects or credit con-
straints is to examine the effect of housing wealth increases on the consumption of
other goods. If households do not consume more of other goods when their home
equity increases, it is reasonable to argue the wealth effect of college is also negligi-
ble. There is a sizeable literature on the relationship between housing wealth and
consumption that lacks consensus. Much of this literature focuses on the United
Kingdom. Using micro-data on consumption combined with aggregate home price
data, Campbell and Cocco (2007) show both regional and national housing price fluc-
tuations have a significant effect on household consumption. Attanasio et al. (2005),
however, argue this relationship is spurious and does not reflect a causal link be-
tween housing wealth and consumption. Using aggregate U.S. state-level data, Case,
Quigley and Shiller (2005) present evidence that states that experienced housing
price increases also experienced consumption increases. Hurst and Stafford (2004),
in an analysis using the 1991-1996 PSID, find households that experience an un-
employment spell between 1991-1994 are significantly more likely to refinance their
home mortgages. Furthermore, households with large loan-to-value ratios (who are
more likely to be liquidity constrained) spend 66% of their tapped equity on current
consumption, while those with small loan-to-value ratios are less likely to use their
housing equity for consumption. Their findings are consistent with households using
their housing equity to finance consumption. In a related analysis using the PSID,
Lehnert (2003) finds sizeable housing wealth elasticities of food consumption, but
only for households with heads aged 52 to 62. Notably, such households constitute
a small portion of my sample. Souleles (2000) provides the most relevant evidence
to this paper on household consumption and educational investment. He employs
data from the Consumer Expenditure Survey to examine how consumption changes
when a household incurs the cost of sending a child to college. In general, he finds
little evidence that nondurable consumption changes with higher education expen-


                                          32
ditures, but there is a small delayed negative effect for all households as well as a
more immediate negative effect for households sending their first child to college.
   To test for evidence of consumption responses to counterfactual home equity
changes in the time period surrounding the college attendance decision, I estimate
equation (9) using four consumption measures in the PSID: total food consumption,
number of automobiles owned, whether the household head took a vacation in the
past year, and the number of vacation weeks taken by the household head in the past
year. The results are presented in Table 8 for the full sample (Panel A), for the sam-
ple living in high housing price and low population growth MSAs (Panel B) and for
all other MSAs (Panel C). In Panel A, the only positive and statistically significant
evidence of a consumption effect of housing wealth is for total food expenditures, but
I find large effects in the 1980s and 1990s as well as in the 2000s. The relationship
between food consumption and housing wealth increases does not exhibit a sharp in-
crease in the 2000s when housing wealth became more liquid, which is when I find a
college attendance response. The estimates in Panel B present a somewhat stronger
argument against wealth effects in my results. Recall that these are the MSAs in
which the college attendance effect is localized, but there is no evidence of increased
consumption due to housing wealth growth in these areas. In contrast, the MSAs in
which there was no college attendance effect are responsible for the observed positive
correlations in the full sample.
   While the estimates in Table 8 are inconsistent with the estimated relationship
between housing wealth and college enrollment being due solely to a wealth effect,
they should be interpreted with care because food, automobiles, and leisure may
have different wealth elasticities than the consumption aspect of college enrollment.
The assumption that college is a pure investment good is prevalent in the literature,
and Table 8 is in line with this assumption. Although it is difficult to separate
wealth effects from credit constraints empirically, the sum total of the evidence is
arguably more consistent with a credit constraint explanation than a wealth effect


                                          33
explanation. More work attempting to separate these two competing hypotheses is
needed in future analyses.



7    Conclusion

With large real increases in the cost of attending college since the 1970s, the as-
sumption of perfect capital markets in models of optimal education investment has
become increasingly problematic. This paper adds to the sizeable literature on credit
constraints and college enrollment by examining the role of housing wealth, which
previous work has ignored. Using the Panel Study of Income Dynamics (PSID), I
construct a repeated cross section of 18-19 year olds tied to their parents’ house-
holds from the years 1980-2005 and examine the effect of the contemporaneous level
and 4-year change in housing prices and wealth on the college enrollment decision.
I conduct the analysis separately by decade and split the sample by income and
wealth into those more likely and less likely to be credit constrained. The reason for
these sample splits is to determine whether my results are being driven by omitted
variables, such as ability, or by local labor market shocks that alter the demand for
high-skilled labor. My results are consistent with the existence of credit constraints
in higher education: in the 2000s, a period when equity extraction and college costs
were growing most rapidly, I find a 1 percent increase in housing equity increases en-
rollment by 0.015 percent. While this elasticity is small, housing prices have increased
by over 360% since 1980, rendering even small effects meaningful. The estimates also
indicate the effects of housing wealth on college enrollment are restricted to those
households most likely to be credit constrained and among those who live in high
housing price growth but low population growth MSAs. The latter result, in particu-
lar, suggests my estimates are not being driven by endogenous housing price changes
at the MSA level.
    I present evidence that among families with 18-22 year old kids, those whose kids



                                          34
are enrolled in college are more likely to have a home equity loan or home equity
line of credit, which is consistent with families tapping their home equity to pay for
college. I argue my results are more consistent with a credit constraint explanation
than with a wealth effect, particularly because I find little evidence of increased
consumption due to housing equity increases in my PSID sample.
   While I lack an experiment that would allow me to identify the causal effect of
housing wealth on enrollment with a perfectly clean research design, the sum total
of the evidence in this paper is difficult to explain without liquidity constraints. The
estimated effects are small, but fluctuations in housing wealth are large, and the
reliance of homeowners on home equity loans is probably growing as the expected
liquidity of their asset has grown.
   These results have particular relevance to current policy as credit markets have
tightened and increases in housing prices have leveled-off or even begun to decline in
many areas of the country. Considering the rising cost of college and the reduction
in family resources caused by these problems in the housing market, it is likely many
families will face increasing constraints in their ability to finance college in the near
future. A simulation of the recent housing market slowdown using my estimates
suggests that the 18.2 percent housing price decline since June 2006 could reduce
college enrollment by about 4 percent. To the extent this will reduce the growth in
college-educated workers, the housing bust can have negative longer-run effects on
economic growth. This consequence of housing market fluctuations largely has been
ignored by policymakers, due primarily to the lack of evidence on the relationship
between college attendance and housing wealth. The central implication of this work
is that college attendance is sensitive to these fluctuations, and future research is
needed on policies that can insulate the training of high-skilled labor from variation
in the housing market.




                                          35
References
[1] Attanasio, Orazio P., Laura Blow, Robert Hamilton, and Andrew Leicester, 2005. “Booms and
    Busts: Consumption, House Prices and Expectations.” IFS Working Paper WP05/24.
[2] Becker, Gary, 1962. “Investment in Human Capital: A Theoretical Analysis.” Journal of Polit-
    ical Economy 70(5): 9–49.
[3] Belley, Philippe and Lance Lochner , 2007. “The Changing Role of Family Income and Ability
    in Determining Educational Achievement.” Journal of Human Capital 1(1): 37–89.
[4] Ben-Porath, Yoram, 1967. “The Production of Human Capital and the Lifecycle of Earnings.”
    Journal of Political Economy 75(4): 352–365.
[5] Bound, John, Michael F. Lovenheim and Sarah E. Turner, 2007. “Understanding the Decrease in
    College Completion Rates and the Increased Time to the Baccalaureate Degree.” PSC Research
    Report No. 07-626, Population Studies Center, University of Michigan.
[6] Brown, Meta, John Karl Scholz and Ananth Seshadri , 2007. “A New Test of Borrowing Con-
    straints for Higher Education.” University of Wisconsin Mimeo.
[7] Brown, Meta, Maurizio Mazzocco, John Karl Scholz and Ananth Seshadri , 2007. “Tied Trans-
    fers.” University of Wisconsin Mimeo.
[8] College Board, 2007. “Trends in Student Aid 2007.” Trends in Higher Education Series.
                             a
[9] Campbell, John Y. and Jo˜o F. Cocco, 2007. “How Do House Prices Affect Consumption?
    Evidence From Micro Data.” Journal of Monetary Economics 54: 591–621.
[10] Cameron, Stephen V. and James J. Heckman, 1998. “Life Cycle Schooling and Dynamic Se-
    lection Bias: Models and Evidence for Five Cohorts of American Males.” Journal of Political
    Economy 106(April): 262–333.
[11] Cameron, Stephen V. and James J. Heckman, 2001. “The Dynamics of Educational Attainment
    for Black, Hispanic, and White Males.” Journal of Political Economy 109(June): 455–499.
[12] Cameron, Stephen V. and Christopher Taber, 2004. “Estimation of Educational Borrowing
    Constraints Using Returns to Schooling” Journal of Political Economy 112(1): 132–182.
[13] Carneiro, Pedro and James J. Heckman, 2002. “The Evidence on Credit Constraints in Post-
    Secondary Schooling.” The Economic Journal 112(482): 705–734.
[14] Case, Karl E., John M. Quigley, and Robert J. Schiller, 2005. “Comparing Wealth Effects: The
    Stock Market versus the Housing Market.” Advances in Macroeconomics 5(1): 2–3.
[15] Chetty, Raj and Adam Szeidl, 2007. “Consumption Commitments and Risk Preferences.”
    Quarterly Journal of Economics 122(2): 831–877.
[16] DeLong, J. Bradford, Claudia Goldin and Laurence Katz, 2003. “Sustaining U.S. Economic
    Growth,” in H. Aaron, J. Lindsay and P. Nivola, eds, Agenda for the Nation, Brookings Insti-
    tution: 17-60.
[17] Dynarski, Susan, 2002. “The Behavioral and Distributional Implications of Aid for College.”
    American Economic Review 92(2): 279–285.
[18] Ellwood, David and Thomas Kane, 2000. “Who is Getting a College Education? Family
    Background and the Growing Gap in Enrollment.” In Securing the Future (Sheldon Danziger
    and Joel Waldfogel, eds.), New York: Russell Sage.
[19] Glaeser, Edward L., Joseph Gyourko and Raven E. Saks, 2005. “Why Have Housing Prices
    Gone Up?” NBER Working Paper No. 11129.

                                              36
[20] Grant, Timothy, 2007. “Need Tuition? Don’t Count on Home Equity Loans.” Pittsburgh Post-
    Gazette, August 17.
[21] Greenspan, Alan and James Kennedy, 2005. “Estimates of Home Mortgage Originations, Re-
    payments and Debt on One-to-Four Family-Residences.” Federal Reserve Board Working Paper,
    Finance and Economics Discussion Series.
[22] Gyourko, Joseph, Christopher Mayer and Todd Sinai, 2006. “Superstar Cities.” Mimeo.
[23] Hurst, Eric and Frank Stafford, 2004. “Home is Where the Equity Is: Mortgage Refinancing
    and Household Consumption.” Journal of Money, Credit and Banking 36(6): 987–1014.
[24] Jappelli, Tullio, J¨rn-Steffen Pischke and Nicholas S. Souleles, 1998. “Testing for Liquidity
                        o
    Constraints in Euler Equations with Complementary Data Sources.” Review of Economics and
    Statistics 80(2): 251–262.
[25] Lehnert, Andreas, 2003. “Housing, Consumption and Credit Constraints.” Mimeo.
[26] Mincer, Jacob, 1997. “The Production of Human Capital and the Life Cycle of Earnings:
    Variations on a Theme.” Journal of Labor Economics 15(1): S26—S47.
[27] Palacios-Huerta, Ignacio, 2003. “An Empirical Analysis of the Risk Properties of Human Cap-
    ital Returns.” American Economic Review 93(3): 948—964.
[28] Plug, Eric and Wim Vijverberg, 2005. “Does Family Income Matter for Schooling Outcomes?
    Using Adoptees as a Natural Experiment.” The Economic Journal 115(506): 879—906.
[29] Runkle, David E., 1991. “Liquidity Constraints and the Permanent Income Hypothesis: Evi-
    dence from Panel Data.” Journal of Monetary Economics 27: 73–98.
[30] Reynolds, Curtis L., 2006. “Where to Attend? Estimates of the Effects of Beginning at a
    Two-year College.” Mimeo.
[31] Reyes, Jessica Wolpaw, 2008. “College Financial Aid Rules and the Allocation of Savings.”
    Education Economics 16(2): 187–189.
[32] Shea, John, 2000. “Does Parents’ Money Matter?” Journal of Public Economics 77(2): 155–
    184.

[33] Souleles, Nicholas S., 2000. “College Tuition and Household Saving and Consumption.” Journal
    of Public Economics 77(2): 185–207.
[34] Zeldes, Stephen P., 1989. “Consumption and Liquidity Constraints: An Empirical Investiga-
    tion” Journal of Political Economy 97(April): 305–346.




                                               37
Table 1: Joint Distributions of Real Family Income Quartiles
         and Real Housing Price and Housing Equity Quar-
         tiles Among Homeowners
                          Panel A: Home Prices
 Family Income                 Home Price Quartile
   Quartile                 1        2      3        4                 Average
       1                 8.08%    3.10%  1.32%    0.52%                13.01%
       2                 9.34%    7.36%  4.24%    1.82%                22.76%
       3                 5.85%    9.98%  9.52%    4.80%                30.14%
       4                 1.78%    4.55%  9.94% 17.83%                  34.09%
    Average              25.05% 24.98% 25.01% 24.96%                    100%
                          Panel B: Home Equity
 Family Income                Home Equity Quartile
   Quartile                 1        2      3        4                 Average
       1                 7.16%    4.35%  2.75%    0.96%                15.22%
       2                 8.45%    6.57%  5.68%    2.61%                23.32%
       3                 6.27%    8.61%  8.35%    5.90%                29.13%
       4                 3.14%    5.49%  8.22% 15.47%                  32.33%
    Average              25.03% 25.03% 25.01% 24.94%                    100%
 1   Source: Reported home prices, remaining mortgage principal, and family in-
     come come from the Panel Study of Income Dynamics (PSID). All quartiles
     are for the period 1980-2005 and are for homeowner households with an 18 or
     19-year old.
 2   The averages at the bottom of each panel are not identical due to a small num-
     ber of observations with missing remaining loan principal information. Home
     equity cannot be calculated for these households and they are dropped from
     the calculations in Panel B but not Panel A.




                                        38
Table 2: Joint Distributions of Real Family Income Quartiles
         and Quartiles of the 4-Year Change in Real Home
         Prices and Home Equity Among Homeowners
              Panel A: Home Price Changes
 Family Income       Quartile of Price Change
   Quartile         1        2        3       4  Average
       1         5.49%    3.85%    3.25%   2.34% 14.93%
       2         6.85%    6.32%    6.05%   3.98% 23.20%
       3         7.30%    8.43%    7.28%   6.21% 29.23%
       4         5.38%    6.54%    8.45% 12.28% 32.64%
    Average      25.01% 25.14% 25.03% 24.81%      100%
             Panel B: Home Equity Changes
 Family Income      Quartile of Equity Change
   Quartile         1        2        3       4  Average
       1         4.73%    4.44%    3.51%   2.20% 14.89%
       2         6.39%    6.83%    5.86%   4.13% 23.21%
       3         7.57%    8.01%    7.90%   5.77% 29.26%
       4         6.32%    5.73%    7.70% 12.89% 32.64%
    Average      25.01% 25.01% 24.98% 25.00%      100%
      Panel C: Counterfactual Home Equity Changes
                    Quartile of Counterfactual
 Family Income           Equity Change
   Quartile         1        2        3       4  Average
       1         7.31%    3.85%    2.75%   1.20% 15.11%
       2         7.85%    6.39%    5.76%   3.28% 23.28%
       3         5.85%    8.64%    8.26%   6.39% 29.13%
       4         4.00%    6.12%    8.26% 14.09% 32.47%
    Average      25.01% 25.00% 25.0% 24.96%       100%
 1   Source: Reported home prices, remaining mortgage principal, and family in-
     come come from the Panel Study of Income Dynamics (PSID). All quartiles
     are for the period 1980-2005 and are for homeowner households with an 18 or
     19-year old.
 2   The averages at the bottom of each panel are not identical due to a small num-
     ber of observations with missing remaining loan principal information. Home
     equity cannot be calculated for these households and they are dropped from
     the calculations in Panels B and C but not Panel A.
 3   The counterfactual change in home equity is defined as the change in housing
     equity expected over the 4-year period if the household accumulated housing
     equity without extracting any of its housing wealth.




                                        39
Table 3: Conditional Ability and Income Gradients from 4 Lon-
         gitudinal Surveys: NLSY79, NLSY97, NLS72, and
         NELS:88
                                       Dependent Variable: Dummy=1 if
                                              Enrolled in College
                                           OLS                   Logit
                                   NLSY79 NLSY97          NLS72     NELS:88
                                    0.132∗∗    0.241∗∗     .           .
 AFQT Quartile 2
                                   (0.026)   (0.025)       .           .
                                    0.324∗∗    0.400∗∗     .           .
 AFQT Quartile 3
                                   (0.028)   (0.026)       .           .
                                    0.549∗∗    0.521∗∗     .           .
 AFQT Quartile 4
                                   (0.029)   (0.026)       .           .
                                     .          .         0.672∗∗    0.638∗∗
 Math Quartile 2
                                     .          .        (0.117)    (0.158)
                                     .          .         1.169∗∗    1.117∗∗
 Math Quartile 3
                                     .          .        (0.131)    (0.166)
                                     .          .         2.052∗∗    2.015∗∗
 Math Quartile 4
                                     .          .        (0.134)    (0.222)
                                     .          .         0.409∗∗    0.107
 Reading Quartile 2
                                     .          .        (0.103)    (0.142)
                                     .          .         0.652∗∗    0.307∗
 Reading Quartile 3
                                     .          .        (0.126)    (0.184)
                                     .          .         0.926∗∗    0.554∗∗
 Reading Quartile 4
                                     .          .        (0.132)    (0.244)
                                    0.023      0.040∗      .           .
 Income Quartile 2
                                   (0.026)   (0.025)       .           .
                                    0.029      0.104∗∗     .           .
 Income Quartile 3
                                   (0.028)   (0.026)       .           .
                                    0.093∗∗    0.161∗∗     .           .
 Income Quartile 4
                                   (0.029)   (0.028)       .           .
                                     .          .        -0.168     -0.053
 Income 6000/20000
                                     .          .        (0.167)    (0.185)
                                     .          .        -0.354∗     0.014
 Income 7500/25000
                                     .          .        (0.183)    (0.215)
                                     .          .         0.069      0.200
 Income 10500/35000
                                     .          .        (0.175)    (0.185)
                                     .          .         0.089      0.335∗
 Income 15000/50000
                                     .          .        (0.192)    (0.184)
                                     .          .         0.505∗∗    0.797∗∗
 Income 15000+/50000+
                                     .          .        (0.226)    (0.192)
 1   The first two results columns are taken from Belley and Lochner (2007), Table
     3. Their college attendance measure is attendance as of age 21. These results
     include controls for sex, age, race, mother’s age, family status during adolescence,
     urban, number of siblings under 18 and mother’s educational attainment.
 2   Results in the second two columns are from the NLS72 and NELS:88 surveys as
     used in Bound, Lovenheim, and Turner (2007). Their college attendance measure
     is attendance within two years of each cohort’s high school graduation, which is
     June 1972 in the NLS72 survey and June 1992 in the NELS:88 survey. These
     regressions include controls for race, mother’s and father’s education and sex.
     Missing values are imputed as discussed in Bound, Lovenheim, and Turner (2007).
 3   Parental income in NLS72 and NELS:88 are given in discrete ranges in both sur-
     veys. I follow Bound, Lovenheim, and Turner (2007) and group the income
     ranges into 6 income categories in each survey that correspond to the same
     real income across surveys using the CPI. In NLS72, the real income ranges are
     less than $3,000, $3001-$6000, $6001-$7500, $7501-$10500, $10501-$15000, and
     greater than $15000. In NELS:88, the real income ranges are less than $10,000,
     $10001-$20000, $20001-$25000, $25001-$35000, $35001-$50000, and greater than
     $50000.




                                           40
      Table 4: Logit Estimates of Probability of Enrollment as a Function of Housing Values
               and Demographic Characteristics by Decade, 1980-2005
                                                                Dependent Variable:      Dummy=1 if Enroll       in College
                 Independent Variable                               (i)     (ii)           (iii)    (iv)               (v)
                                                                 0.363∗∗  0.393∗∗         0.409∗∗  0.436∗∗          0.388∗∗
                  Homeowner Dummy
                                                                (0.122)  (0.120)         (0.119)  (0.119)         (0.118)
                                                                 0.004     .               .        .                .
           I(1980s)*House Value ($10,000)
                                                                (0.007)    .               .        .                .
                                                                 0.006     .               .        .                .
           I(1990s)*House Value ($10,000)
                                                                (0.005)    .               .        .                .
                                                                 0.010∗    .               .        .                .
           I(2000s)*House Value ($10,000)
                                                                (0.005)    .               .        .                .
                                                                  .       0.003            .        .                .
         I(1980s)*Housing Equity ($10,000)
                                                                  .      (0.007)           .        .                .
                                                                  .       0.003            .        .                .
         I(1990s)*Housing Equity ($10,000)
                                                                  .      (0.006)           .        .                .
                                                                  .       0.007            .        .                .
         I(2000s)*Housing Equity ($10,000)
                                                                  .      (0.006)           .        .                .
                                                                  .        .             -0.002     .                .
     I(1980s)*Housing Price Change ($10,000)
                                                                  .        .             (0.008)    .                .
                                                                  .        .              0.005     .                .
     I(1990s)*Housing Price Change ($10,000)
                                                                  .        .             (0.008)    .                .
                                                                  .        .              0.019∗∗   .                .
     I(2000s)*Housing Price Change ($10,000)
                                                                  .        .             (0.008)    .                .
                                                                  .        .               .      -0.007             .
     I(1980s)*Housing Equity Change ($10,000)
                                                                  .        .               .      (0.009)            .
                                                                  .        .               .      -0.004             .
     I(1990s)*Housing Equity Change ($10,000)
                                                                  .        .               .      (0.009)            .
                                                                  .        .               .       0.014∗            .
     I(2000s)*Housing Equity Change ($10,000)
                                                                  .        .               .      (0.008)            .
           I(1980s)*Counterfactual Housing                        .        .               .        .               0.002
               Equity Change ($10,000)                            .        .               .        .             (0.009)
           I(1990s)*Counterfactual Housing                        .        .               .        .               0.007
               Equity Change ($10,000)                            .        .               .        .             (0.009)
           I(2000s)*Counterfactual Housing                        .        .               .        .               0.028∗∗
               Equity Change ($10,000)                            .        .               .        .             (0.009)
                                                                 0.017∗∗  0.017∗          0.019∗   0.020∗∗          0.018∗
       I(1980s)*Real Family Income ($10,000)
                                                                (0.010)  (0.010)         (0.010)  (0.010)         (0.010)
                                                                 0.039∗∗  0.042∗∗         0.043∗∗  0.045∗∗          0.043∗∗
       I(1990s)*Real Family Income ($10,000)
                                                                (0.015)  (0.015)         (0.014)  (0.014)         (0.015)
                                                                 0.015    0.020∗∗         0.017∗   0.018∗           0.013
       I(2000s)*Real Family Income ($10,000)
                                                                (0.010)  (0.010)         (0.009)  (0.010)         (0.009)
                                                                 0.150    0.170∗          0.180∗   0.174∗           0.171∗
I(1980s)*Real Yearly Mortgage Payments ($10,000)
                                                                (0.104)  (0.097)         (0.098)  (0.098)         (0.100)
                                                                -0.019∗∗ -0.018∗∗        -0.019∗∗ -0.019∗∗         -0.019∗∗
I(1990s)*Real Yearly Mortgage Payments ($10,000)
                                                                (0.008)  (0.008)         (0.008)  (0.008)         (0.008)
                                                                 0.005    0.006           0.005    0.006            0.005
I(2000s)*Real Yearly Mortgage Payments ($10,000)
                                                                (0.006)  (0.006)         (0.007)  (0.007)         (0.007)
1 Source: Panel Study of Income Dynamics repeated cross-section of 18-19 year olds as described in the text.
2 All financial variables are in real 2007 $10,000, adjusted using the CPI. All models include year fixed effects as well as
  controls for household head’s education level, age, sex and marital status, respondent’s age, sex and race, and the
  number of other dependents in the household. The regressions are weighted by the family weights in the PSID.
3 Robust standard errors are in parentheses: * indicates significance at the 10 percent level and ** indicates significance
  at the 5 percent level.




                                                               41
      Table 5: Logit Estimates of Probability of Enrollment as a Function of Housing Values
               and Demographic Characteristics by Decade, Sample Splits Based on Income
               and Wealth Levels
                                                         Dependent Variable: Dummy=1 if Enroll in College
                                                                                           Real Family Income
                                                    Family Income       CF Home Equity       CF Home Equity
                                                  Under       Over      Under     Over      Under         Over
                                                 $125,000 $125,000 $125,000 $125,000          1.5          1.5
          Independent Variable                      (i)        (ii)       (iii)    (iv)       (v)         (vi)
                                                  0.288∗∗    0.418      0.307∗∗  0.808∗∗    0.410∗∗     0.341∗
           Homeowner Dummy
                                                 (0.125)    (0.467)    (0.114)  (0.163)    (0.115)    (0.141)
    I(1980s)*Counterfactual Housing              -0.002     -0.001     -0.023   -0.008      0.001      -0.077
        Equity Change ($10,000)                  (0.010)    (0.016)    (0.021)  (0.010)    (0.009)    (0.047)
    I(1990s)*Counterfactual Housing               0.008      0.001      0.024   -0.002     -0.002       0.055∗
        Equity Change ($10,000)                  (0.010)    (0.012)    (0.020)  (0.008)    (0.008)    (0.030)
    I(2000s)*Counterfactual Housing               0.039∗∗    0.015      0.057∗∗  0.012      0.029∗∗ -0.021
        Equity Change ($10,000)                  (0.015)    (0.009)    (0.027)  (0.009)    (0.011)    (0.046)
                                                  0.065∗∗    0.016      0.044∗∗  0.005      0.027∗      0.029∗∗
I(1980s)*Real Family Income ($10,000)
                                                 (0.023)    (0.013)    (0.017)  (0.006)    (0.015)    (0.014)
                                                  0.106∗∗    0.002      0.060∗∗  0.024      0.079∗∗     0.014
I(1990s)*Real Family Income ($10,000)
                                                 (0.022)    (0.013)    (0.018)  (0.015)    (0.017)    (0.014)
                                                  0.058∗∗    0.006      0.014    0.011      0.021       0.012
I(2000s)*Real Family Income ($10,000)
                                                 (0.028)    (0.005)    (0.015)  (0.008)    (0.018)    (0.013)
        I(1980s)*Real Yearly                      0.289∗∗   -0.124     -0.074    0.153      0.159      -0.156
     Mortgage Payments ($10,000)                 (0.122)    (0.215)    (0.166)  (0.126)    (0.108)    (0.292)
        I(1990s)*Real Yearly                     -0.035∗∗    0.024     -0.026∗  -0.019∗    -0.034       0.023
     Mortgage Payments ($10,000)                 (0.009)    (0.035)    (0.013)  (0.010)    (0.010)    (0.027)
        I(2000s)*Real Yearly                     -0.005      0.008      0.037∗∗ -0.003      0.000       0.320
     Mortgage Payments ($10,000)                 (0.010)    (0.010)    (0.017)  (0.007)    (0.007)    (0.210)
       Number of Observations                      6,386       818       6,162    3,423     5,633        3,838
1 Source: Panel Study of Income Dynamics repeated cross-section of 18-19 year olds as described in the text.
2 All financial variables are in real 2007 $10,000, adjusted using the CPI. All models include year fixed effects as well as
  controls for household head’s education level, age, sex and marital status, respondent’s age, sex and race, and the
  number of other dependents in the household. The regressions are weighted by the family weights in the PSID.
3 The estimation sample in each column includes all renters.
4 Robust standard errors are in parentheses: * indicates significance at the 10 percent level and ** indicates
  significance at the 5 percent level.




                                                               42
Table 6: Logit Estimates of Probability of Enrollment as a Function of Housing Values
         and Demographic Characteristics by Decade, by MSA-Growth Type and Using
         Predicted MSA-Level Price Growth
                                                       Dependent Variable: Dummy=1 if Enroll in              College
                                                                 MSA Growth Type                             Predicted
                                                          1990-2005               2000-2005                    MSA-
                                                     Low Pop        All     Low Pop         All                Level
                                                    High Price    Other    High Price     Other               Growth
             Independent Variable                       (i)         (ii)       (iii)       (iv)                  (v)
                                                     0.296∗∗      0.422∗∗   0.318∗        0.420∗∗             0.347∗∗
             Homeowner Dummy
                                                    (0.151)      (0.201)   (0.198)       (0.172)             (0.088)
      I(1980s)*Counterfactual Housing               -0.015        0.019    -0.012         0.021               0.014
          Equity Change ($10,000)                   (0.012)      (0.012)   (0.010)       (0.015)             (0.019)
      I(1990s)*Counterfactual Housing               -0.001        0.004     0.004        -0.003              -0.009
          Equity Change ($10,000)                   (0.015)      (0.010)   (0.018)       (0.008)             (0.022)
      I(2000s)*Counterfactual Housing                0.053∗∗      0.008     0.043∗∗       0.009               0.035∗∗
          Equity Change ($10,000)                   (0.022)      (0.008)   (0.021)       (0.009)             (0.017)
                                                     0.020∗∗      0.029∗    0.023∗∗       0.030∗              0.022∗∗
 I(1980s)*Real Family Income ($10,000)
                                                    (0.008)      (0.018)   (0.009)       (0.017)             (0.011)
                                                     0.051∗∗      0.044∗∗   0.077∗∗       0.034∗∗             0.051∗∗
 I(1990s)*Real Family Income ($10,000)
                                                    (0.019)      (0.020)   (0.023)       (0.016)             (0.015)
                                                     0.015        0.011     0.042         0.011               0.019∗
 I(2000s)*Real Family Income ($10,000)
                                                    (0.016)      (0.011)   (0.031)       (0.009)             (0.011)
           I(1980s)*Real Yearly                      0.340∗∗      0.116     0.240         0.129               0.159∗
        Mortgage Payments ($10,000)                 (0.167)      (0.099)   (0.172)       (0.100)             (0.085)
           I(1990s)*Real Yearly                      0.002        0.017     0.001         0.018               0.006
        Mortgage Payments ($10,000)                 (0.013)      (0.019)   (0.011)       (0.017)             (0.012)
           I(2000s)*Real Yearly                      0.027        0.004    -0.001         0.043               0.011
        Mortgage Payments ($10,000)                 (0.006)      (0.011)   (0.012)       (0.029)             (0.007)
 1   Source: Panel Study of Income Dynamics repeated cross section of 18-19 year olds as described in the text. The
     estimation sample includes only those living in an identified MSA who do not move across MSAs between time t-4
     and t.
 2   All financial variables are in real 2007 $10,000, adjusted using the CPI. All models include year fixed effects as well
     as controls for household head’s education level, age, sex and marital status, respondent’s age, sex and race, and the
     number of other dependents in the household. The regressions are weighted by the family weights in the PSID.
 3   Low population growth MSAs are those with 1990-2005 or 2000-2005 annualized population growth below the median.
     High price growth MSAs are those with above median 1990-2005 or 2000-2005 annualized housing price growth rates,
     calculated from the MSA-level CMHPI. All other MSAs are all MSAs with above-median population or below-median
     housing price growth.
 4   Standard errors are in parentheses and are clustered at the MSA-level: * indicates significance at the 10 percent level
     and ** indicates significance at the 5 percent level.




                                                         43
Table 7: Logit Estimates of Probability of Hav-
         ing a Home Equity Loan or Home Equity
         Line of Credit Among Homeowners with
         an 18 to 22-year Old in the Household,
         1994-2005
                                     Dependent Variable:
                                     Dummy=1 if Have a
                                       Home Equity Loan
 Independent Variable              (i)        (ii)     (iii)
                                0.213∗∗     0.211∗∗   0.219∗∗
 Dependent in College
                               (0.108)     (0.108)   (0.108)
                                 .         -0.000    -0.003
 Real Family Income
                                 .         (0.001)   (0.005)
 Real Family Wealth              .         -0.002    -0.002
 Excluding Housing               .         (0.001)   (0.001)
                                 .           .        0.005∗
     Real Home Price
                                 .           .       (0.003)
                               -8.566∗∗ -8.676∗∗ -8.393∗∗
         Constant
                               (1.769)     (1.790)   (1.748)
             R2                 0.053       0.052     0.052
 1   Source: Panel Study of Income Dynamics repeated cross section
     of 18-22 year olds whose family owns a home as described in
     the text.
 2   All financial variables are in real 2007 $10,000, adjusted using
     the CPI. Demographic controls include indicators for household
     head being white, father’s and mother’s education level dummy
     variables, number of other dependents in household between 0-
     5, 6-10, 11-17, 18-22, 22+, age of household head, and marital
     status of household head. The regressions are weighted by the
     family weights in the PSID.
 3   Robust standard errors are in parentheses: * indicates signif-
     icance at the 10 percent level and ** indicates significance at
     the 5 percent level.




                                 44
Table 8: Estimates of The Relationship Between Housing Wealth Changes and Differ-
         ent Measures of Consumption by Decade

                                      Panel A: Full Sample
                                                             Dependent Variable:
                                                                                    Number
                                            Total       Number         Take             of
                                             Food           of           a          Vacation
                                         Expenditure Automobiles Vacation?           Weeks
                                             OLS          OLS         Logit           OLS
           Independent Variable               (i)          (ii)        (iii)           (iv)
                                                             ∗∗
                                          306.90       0.455         0.388∗∗         0.636∗∗
            Homeowner Dummy
                                         (224.02)     (0.182)       (0.131)         (0.118)
     I(1980s)*Counterfactual Housing      113.28∗∗     0.010         0.009           0.003
         Equity Change ($10,000)          (31.96)     (0.005)       (0.016)         (0.009)
     I(1990s)*Counterfactual Housing       37.66∗     -0.001         0.004          -0.005
         Equity Change ($10,000)          (19.95)     (0.007)       (0.020)         (0.007)
     I(2000s)*Counterfactual Housing       73.48∗∗     0.005         0.010          -0.003
         Equity Change ($10,000)          (19.04)     (0.004)       (0.014)         (0.006)
                                           84.53       0.013         0.022           0.014
  I(1980s)*Real Family Income ($10,000)
                                          (59.72)     (0.011)       (0.071)         (0.007)
                                          172.39∗∗     0.018∗∗       0.107∗∗         0.213∗
  I(1990s)*Real Family Income ($10,000)
                                          (42.03)     (0.009)       (0.032)         (0.114)
                                           54.56∗∗     0.000         0.046           0.003
  I(2000s)*Real Family Income ($10,000)
                                          (14.12)     (0.005)       (0.050)         (0.006)
             Panel B: Above Median Housing and Below Median Pop Growth (90-05)
                                                       Dependent Variable:
                                                                                    Number
                                               Total       Number         Take          of
                                               Food            of           a       Vacation
                                           Expenditure Automobiles Vacation?         Weeks
                                                OLS          OLS         Logit        OLS
          Independent Variable                   (i)          (ii)        (iii)        (iv)
                                                                ∗∗
                                            501.59        0.982         0.162        0.505∗∗
          Homeowner Dummy
                                           (397.14)      (0.256)       (0.328)      (0.193)
    I(1980s)*Counterfactual Housing         124.56∗∗     -0.004         0.006        0.004
        Equity Change ($10,000)              (25.81)     (0.012)       (0.027)      (0.021)
    I(1990s)*Counterfactual Housing           11.51      -0.004        -0.004       -0.005
        Equity Change ($10,000)              (26.02)     (0.008)       (0.024)      (0.009)
    I(2000s)*Counterfactual Housing           -3.72      -0.022∗∗       0.013       -0.025
        Equity Change ($10,000)              (45.79)     (0.006)       (0.031)      (0.015)
                                            107.61∗∗      0.037∗∗       0.060        0.030
  I(1980s)*Real Family Income ($10,000)
                                             (44.12)     (0.013)       (0.057)      (0.028)
                                            204.69∗∗      0.068∗        0.189∗∗      0.079∗∗
  I(1990s)*Real Family Income ($10,000)
                                             (37.43)     (0.009)       (0.060)      (0.025)
                                            185.54∗∗      0.036∗∗       0.126∗∗      0.022
  I(2000s)*Real Family Income ($10,000)
                                             (66.22)     (0.013)       (0.049)      (0.022)
                                  Panel   C: All Other MSAs
                                                          Dependent Variable:
                                                                                    Number
                                              Total           Number         Take      of
                                              Food              of            a     Vacation


                                              45
                                                 Expenditure      Automobiles       Vacation?       Weeks
                                                     OLS              OLS             Logit          OLS
          Independent Variable                        (i)              (ii)            (iii)          (iv)
                                                  557.41∗∗         0.161             0.413∗∗        0.770∗∗
           Homeowner Dummy
                                                 (276.34)         (0.120)           (0.185)        (0.183)
    I(1980s)*Counterfactual Housing                83.26∗∗         0.021             0.033∗∗        0.006
        Equity Change ($10,000)                   (26.62)         (0.014)           (0.016)        (0.012)
    I(1990s)*Counterfactual Housing                16.96           0.009             0.040         -0.012
        Equity Change ($10,000)                   (35.27)         (0.028)           (0.040)        (0.010)
    I(2000s)*Counterfactual Housing                63.77∗∗         0.012∗∗           0.017          0.003
        Equity Change ($10,000)                   (18.46)         (0.004)           (0.020)        (0.010)
                                                   45.19           0.001            -0.012         -0.005
I(1980s)*Real Family Income ($10,000)
                                                  (48.88)         (0.007)           (0.011)        (0.009)
                                                  156.33∗∗         0.010             0.156∗∗        0.013
I(1990s)*Real Family Income ($10,000)
                                                  (64.03)         (0.010)           (0.048)        (0.008)
                                                   48.24∗∗        -0.002             0.021         -0.000
I(2000s)*Real Family Income ($10,000)
                                                  (12.02)         (0.005)           (0.067)        (0.007)
1 Source: Panel Study of Income Dynamics repeated cross section of 18-19 year olds as described in
  the text. The estimation sample in Panels B and C includes only those living in an identified MSA who
  do not move across MSAs between time t-4 and t.
2 All financial variables are in real 2007 $10,000, adjusted using the CPI. All models include year fixed
  effects as well as controls for household head’s education level, age, sex and marital status, respondent’s
  age, sex and race, and the number of other dependents in the household. The regressions are weighted by
  the family weights in the PSID.
3 Low population growth MSAs are those with 1990-2005 annualized population growth below the median.
  High price growth MSAs are those with above median 1990-2005 annualized housing price growth rates,
  calculated from the MSA-level CMHPI. All other MSAs are all MSAs with above-median population or
  below-median housing price growth.
4 Estimates for total food expenditures exclude 1988 and estimates for number of automobiles exclude years
  1988-1996. The PSID did not collect consumption information on these goods in these years.
5 Standard errors are in parentheses and are clustered at the MSA-level: * indicates significance at the 10
  percent level and ** indicates significance at the 5 percent level.




                                                     46
                   Figure 1: Education Investment Decisions and the Cost of Funds
                   r

               r                                                        MCOF’
                                                                                                             MCOF


                                                                                              ACOF= ri
                ri
         Yci + p
     ln(          )
         Y0 + p                                                                                              ACOF’ = ri '
rm =
          T r'
                 i
              r




                   r




                                                                                  C             $
                                                                                                    Y i +p
                                                                                              ln( Yc +p )
                                                                                                   0
    The figure depicts the optimal education investment choice for an agent for whom rm =          T
                                                                                                        . ACOF is the
    average cost of funds and MCOF is the marginal cost of funds. C is the total demand for funds and rm is the
    market interest rate. r1 refers to the interest rate on government subsidized loans, r2 refers to the interest rate
    on home equity, and r3 refers to the interest rate on private loans. Because the investment decision is binary,
                                                                                                                Y i +p
                                                                                                             ln( Yc +p )
                                                                                                                  0
    the individual invests in college if the intersection of the ACOF curve and C is less than or equal to       T
                                                                                                                           ,
    which equals rm in the figure.




                                                          47
Figure 2: Comparison of Home Price Indices Constructed from Self-Reported PSID
          Home Prices and the Published CMHPI
                  500

                                          PSID−Median               PSID−Mean
                                          CMHPI
                             400
    Home Price Index (1980=100)
  100     200     0  300




                                   1980     1985        1990               1995     2000               2005
                                                                    Year

    Source: The PSID indices are the respective reported mean and median home prices in each year for the full
    PSID sample, rescaled such that year 1980=100. The Conventional Mortgage Housing Price Index (CMHPI) is
    reported by Fannie Mae for single-family repeat sale homes and appraisals.




                                                               48
Figure 3: Homeownership Rates Among Families with 18-19 Year Olds Enrolled in
          College, 1977-2003
             .95     .9
  Homeownership Rate
        .85  .8




                                                  Full Sample                2 Year
                                                  4 Year
             .75




                          1977   1980   1983   1986   1989 1992         1995      1998       2001       2004
                                                         Year

   Source: College enrollment rates, college types, and homeownership status are taken from the October Current
   Population Survey.




                                                       49
  Figure 4: College Enrollement Rates by Homeownership Status, 1977-2003

                                                                   Panel A: Full Sample




                             .7        .6
                College Enrollment Rate
                .2    .3   .4.1  .5




                                                                     Homeowners               Renters
                             0




                                            1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
                                                                               Year


                                                                Panel B: 4−Year Students
                             .7
                             .6
                College Enrollment Rate
                 .2    .3     .4
                             .1     .5




                                                                     Homeowners               Renters
                             0




                                            1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
                                                                               Year


                                                                Panel C: 2−Year Students
                             .7
                             .6
                College Enrollment Rate
                 .2    .3     .4
                             .1     .5




                                                                     Homeowners               Renters
                             0




                                            1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005
                                                                               Year

Source: College enrollment rates, college types, and homeownership status are taken from the October Current
Population Survey.




                                                                            50
Figure 5: Trends in Average Home Prices, Housing Equity and Counterfactual Hous-
          ing Equity
            200000
                 150000
  Real 2007 Dollars
     100000 50,000
            0




                          1980 1982 1984 1986 1988 1990 1992 1994 1996 1999 2001 2003 2005

                                 Home Prices              Home Equity                     CF Home Equity

    Source: Author’s calculations from the PSID sample of home-owning families with 18 or 19-year olds in each
    survey year as described in the text. All means are deflated to real 2007 dollars using the CPI-U. Counterfactual
    housing equity is the equity each homeowner would have accumulated if he did not tap any equity in his home
    in the four years prior to the survey year.




                                                        51
Figure 6: Ratio of Extracted Home Equity and Median Home Prices to Average per-
          Capita Income
                        10
                        9
  Home Prices to Per Capita Income
                                8
     Ratio of Home Equity and

   2   3    4    5    6 1  7




                                                   Extracted Home Equity             Home Prices
                        0




                                     1990   1992    1994      1996          1998   2000       2002          2004
                                                                     Year

   Sources: Estimates of gross equity extraction as a ratio of disposable income are taken from Table 1 in Greenspan
   and Kennedy (2005). Median home prices come from the Office of Federal Housing Enterprise Oversight, and
   average per-capita income at the state level comes from “personal income” estimates calculated by the U.S.
   Bureau of Labor Statistics.




                                                               52
Figure 7: Trends in College Tuition, the Cost of College, and Federal Financial Aid,
          1980-2005

                                                    Panel A: Average Yearly Tuition and Fees at U.S.
                                                     Undergraduate Colleges and Universities as a
                                                          Percent of Real Per−Capita Income
                                80                      Public 4 Year                Private 4 Year
                                                        Public 2 Year
                                60
                   Percent
                     40         20
                                0




                                             1975       1980        1985          1990            1995     2000           2005
                                                                                  Year


                                             Panel B: Trends in Cumulative Total Cost of Public 4−year
                                                     Attendance and Federal Financial Aid for
                                                      Students Attending College for 4 Years
                              60000




                                                        Net Public 4−Year Cost                 Total Public 4−Year Cost
                                                        Maximum Pell + Stafford
                                     50000
                      Real 2007 Dollars
                   30000   40000
                              20000




                                             1980         1985           1990              1995          2000             2005
                                                                                  Year

  Sources: Yearly tuition and fees in Panel A are taken from the College Board’s 2005 Trends in Pricing com-
  pilation. Average per-capita income comes from “personal income” estimates calculated by the U.S. Bureau
  of Labor Statistics. Net and total 4-year costs in Panel B are taken from the College Board’s 2007 Trends
  in Pricing compilation. Maximum Pell Grant and Stafford Loan amounts are taken from the FinAid web-
  site: http://www.finaid.org/loans/historicallimits.phtml. Calculations of four-year cumulative Stafford values
  are completed by summing the value of the maximum freshman award in matriculating year, the maximum
  sophomore award in the second year, the maximum junior award in the third year, and the maximum senior
  award in fourth year. All other values in Panel B represent sums over the four years starting in each given year
  on the x-axis. All costs in Panel B include tuition, fees, room and board.




                                                                                53
Figure 8: Changes in HELOC, Stafford, and PLUS Loan Interest Rates, 1980-2005
               20
                                            HELOC Interest Rate                  Stafford Interest Rate
                                            PLUS Interest Rate
                         15
 Interest Rate (percent)
           10  5
               0




                              1980   1985          1990               1995             2000               2005
                                                               Year

   Sources: HELOC interest rates are indexed to the prime rate using a constant markup of 0.39 percent over prime.
   The 0.39 percent markup is the average markup over prime among those with home equity loans in the 2004
   Survey of Consumer Finances. Stafford Loan interest rates are based on the 91-day rate from the last Treasury
   auction in May plus a constant markup equal to 3.25 percent prior to 1992, to 3.1 percent between 1992 and
   1997, and to 2.3 percent between 1998 and 2005. Stafford Loan interest rates were subject to caps of 10 percent
   prior to 1992, of 9 percent between 1992 and 1993, and of 8.25 percent between 1994 and 2005. PLUS interest
   rates are based on average one-year constant maturity Treasury yield (CMT) for the last calendar week in May
   plus a constant markup of 3.25 percent prior to 1992 and of 3.1 percent between 1992 and 2005. PLUS loans
   were capped at 12 percent prior to 1992, at 10 percent between 1992 and 1993, and at 9 percent between 1994
   and 2005.




                                                          54
Table A-1: MSAs With High Housing Price and Low Population Growth Rates
                               Annualized Growth Time Period
                    1990-2005                                    2000-2005
              MSA Name                   State             MSA Name                                  State
                 Fresno                   CA               Los Angeles                                CA
              Los Angeles                 CA                 Oakland                                  CA
                Oakland                   CA              San Francisco                               CA
               San Diego                  CA                 San Jose                                 CA
             San Francisco                CA   Santa Barbara-Santa Maria-Lompoc                       CA
                San Jose                  CA               Santa Rosa                                 CA
 Santa Barbara-Santa Maria-Lompoc         CA          Vallejo-Fairfield-Napa                           CA
              Santa Rosa                  CA                Bridgeport                                CT
                Ventura                   CA                 Danbury                                  CT
         Vallejo-Fairfield-Napa            CA                 Hartford                                 CT
               Bridgeport                 CT               New Haven                                  CT
                Danbury                   CT          New London-Norwich                              CT
                 Macon                    GA                  Macon                                   GA
    Davenport-Moline-Rock Island          IA                 Honolulu                                 HI
                Dubuque                   IA          Waterloo-Cedar Falls                            IA
                Chicago                   IL           Champaign-Urbana                               IL
              Peoria-Pekin                IL                 Chicago                                  IL
                Lafayette                 LA               New Orleans                                LA
              New Orleans                 LA         Shreveport-Bossier City                          LA
               Fitchburg                  MA                  Boston                                  MA
               Baltimore                  MD          Barnstable-Yarmouth                             MA
               Ann Arbor                  MI                 Brockton                                 MA
                  Flint                   MI                   Lowell                                 MA
                 Detroit                  MI              New Bedford                                 MA
         Lansing-East Lansing             MI                Worcester                                 MA
                St. Louis                 MO                Baltimore                                 MD
           Duluth-Superior                MN             Duluth-Superior                              MN
      Biloxi-Gulfport-Pascagoula          MS                 St. Louis                                MO
                Bismarck                  ND             Bergen-Passaic                               NJ
            Bergen-Passaic                NJ               Jersey City                                NJ
              Jersey City                 NJ      Middlesex-Somerset-Hunterdon                        NJ
   Middlesex-Somerset-Hunterdon           NJ                  Newark                                  NJ
                 Newark                   NJ               Ocean City                                 NJ
                 Trenton                  NJ                  Trenton                                 NJ
                 Nassau                   NY                  Nassau                                  NY
Newburgh-Middletown-Poughkeepsie          NY                New York                                  NY
               New York                   NY                 Syracuse                                 NY
                  Akron                   OH            Eugene-Springfield                             OR
           Canton-Massillon               OH       Harrisburg-Lebanon-Carlisle                        PA
          Eugene-Springfield               OR                Lancaster                                 PA
               Charleston                 SC               Philadelphia                               PA
              Philadelphia                PA     Scranton–Wilkes-Barre–Hazleton                       PA
         Salt Lake City-Ogden             UT                Providence                                RI
Virginia Beach-Norfolk-Newport News       VA                Lynchburg                                 VA
               Eau Claire                 WI                Burlington                                VT
               Janesville                 WI                Bremerton                                 WA
               Milwaukee                  WI                Milwaukee                                 WI
               Sheboygan                  WI                  Racine                                  WI
                 Racine                   WI
                 Casper                  WY
               Cheyenne                  WY
1   The table shows MSAs with annualized population growth below median and housing price growth above
    median. The first two columns use annualized growth rates between 1990 and 2005, and the second two
    columns use annualized growth rates between 2000 and 2005.
2   Annualized housing price growth is calculated using the MSA-level CMHPI, and annualized population growth
    is calculated using U.S. Census Bureau MSA-level population estimates.


                                                     55

				
DOCUMENT INFO