The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
STRUCTURAL CHANGE IN 11IE MORTGAGE MARKET AND THE
PROPENSITY TO REFINANCE
Paul Bennett, Richard Peach, and Stavros Peristiani
Federal Reserve Bank of New York
Research Paper No. 9736
Novembi:r 1997
This paper is being circulated for purposes of discussion and comment.
The views expressed are those of the author and do not necessarily reflect those
of the Federal Reserve Bank of New York of the Federal Reserve System.
Single copies are available on request to:
Public Information Department
Federal Reserve Bank of New York
NewYork,NY10045
STRUCTURAL CHANGE IN THE MORTGAGE MARKET AND THE
PROPENSITY TO REFINANCE
Paul Bennett, Richard Peach, Stavros Peristiani *
November 19, 1997
Abstract: We hypothesize that the intrinsic benefit required to trigger a refinancing has
become smaller, due to a combination of technological, regulatory, and structural changes
that have made mortgage origination more competitive and more efficient. To test this
hypothesis, we estimate an empirical hazard model of loan survival for two subperiods,
using a database that allows us to carefully control for homeowners' credit ratings, equity,
loan size, and measurable transaction costs. Our findings strongly confirm that credit
ratings and home equity have significant effects on refinancing probability. In addition, we
provide evidence that homeowners postpone refinancing in the face of increased interest rate
volatility, consistent with option value theory. Finally, our results clearly support the
hypothesis that structural change in the mortgage market has increased homeowners'
propensity to refinance.
Send correspondence to: Paul Bennett, 33 Liberty Street, Main 3, Federal Reserve Bank of New
York, New York, NY 10045. Tel: (212)-720-5647; Fax: (212)-720-1291.
* The authors thank Fred Furlong for helpful comments, Dibora Amanuel and Reagan Murray
for valuable research assistance. The views expressed in this paper are those of the authors and
do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal
Reserve System.
I.
INTRODUCTION
A primary consideration in the pricing of residential mortgage loan assets is prepayment
risk--the premature or unscheduled return of principal to investors when homeowners move,
refinance, or default. Prepayment speeds have increased significantly in the 1990s relative to the
1980s, and this increase cannot be explained by changes in the independent variables normally
used in modeling prepayment behavior. As seen in Figure I, prepayments speeds for moderately
seasoned and seasoned Fannie Mae mortgage backed securities (MBS) backed by 30 year fixed
rate mortgages were substantially higher in the 1990s refinancing waves than was the case in the
1986-1987 wave, despite the fact that the decline in mortgage interest rates from 1983-1984 to
early 1987 was somewhat greater than the decline that occurred from 1990 to late 1993. Rather,
it appears that the quantitative relationships between prepayments--particularly refinancings--and
those explanatory variables have changed in ways that make prepayment more likely, all else
equal.
Significant changes on both the supply and demand sides of the mortgage market likely
contributed to this rise of prepayment speeds by reducing transactions costs or "frictions"
associated with obtaining a new loan. Indeed, over the past twenty-five years the U.S. housing
finance system has undergone a fundamental restructuring. As seen in Fj_gure 2, in the 1970s the
provision of long-term financing to homeownc:rs was dominated by portfolio lenders, primarily
thrift institutions. Due to a complex combination of economic and regulatory changes as well as
financial and technological innovations, today mortgage lending is dominated by mortgage
1
bankers/brokers and the process of securitization. 1• 2 An important distinction between these two
systems is that the former, often referred to as the New Deal system, effectively limited
competition among lenders. 3 In contrast, the modern system has eliminated most geographic
and financial barriers to entry and so is extremely competitive (see Weicher (1994)).4
Increased competition in the primary mortgage market along with improvements in
information processing technology have lowered the explicit, financial transactions costs
associated with obtaining a mortgage, as reflected in the secular decline in points and fees
(Figure 3). Nonfinancial transactions costs were also reduced in the form of shortened time
periods from application to approval and approval to closing and lending programs with
substantially reduced financial documentation in the application process. Furthermore, advances
in computer technology have enabled lenders to quickly and cheaply identify and contact
mortgagors with interest rates above prevailing market rates, thereby disseminating information
about refinancing opportunities more quickly and broadly than occurred in the past. In addition,
1
Unlike thrifts, mongage bankers are not depository institutions. They fund mongages through assoned
forms of shon-term borrowing, often termed a warehouse line of credit, and then sell the loans (for cash or "swap"
them for mongage backed securities) in the secondary mongage market. In most cases the loans are sold "servicing
retained", where the originating mongage banker then services the loan (collects monthly payments for principal,
interest, propeny taxes, and propeny insurance (PITI) and distributes those funds, net of a servicing fee, to the
appropriate parties). Mongage brokers do much of the work of originating a loan but typically do not have a
warehouse line and so must arrange for another lender to fund the mongage at the closing table.
2
Note, however, that many of the nation's largest mongage bankers are owned by bank holding companies.
3
Under the ''New Deal system" Savings & Loan institutions accepted local time deposits and made long
term mongage loans on homes located within 50 miles of their home offices (100 miles after 1964). The Federal
Home Loan Bank Board regulated and supervised the S&L's (established reserve requirements), the Federal Home
Loan Banks served as a discount window, and deposits were insured by the Federal Savings and Loan Insurance
Corporation (FSLIC). Regulation Q established maximum interest rates on deposits, giving thrifts a 25 basis point
higher ceiling than commercial banks.
4sradley, Gabriel, and Wohar (1995) examine the declining role of thrifts. They find that thrifts
significantly influenced the interest rate spread between mongages and treasuries during 1972 to 1982, but little after
that.
2
the growth of the subprime mortgage market established a flow of credit to borrowers unable to
meet the underwriting criteria of the government sponsored enterprises (GSEs). Reinforcing
these developments on the supply side of the primary market, homeowners have likely become
more financially savvy, increasing their propensity to refinance for a given set of measurable
incentives.
This paper presents a formal test of the hypothesis that the propensity to refinance has
increased over time. Conducting such a test rnpresents a considerable challenge. Recent
research has demonstrated quite convincingly that prepayment behavior, particularly refinancing
behavior, is strongly influenced by individual borrower and property characteristics. For
example, Peristiani et al ( 1997) find that in addition to changes in interest rates and transaction
costs, individual homeowners' equity and credit histories play an important role in determining
the probability that a mortgage will be refinanced. Accordingly, a convincing empirical test of
this hypothesis must be based on loan level data that captures these individual borrower
characteristics and which identifies the reason for loan prepayment. Furthermore, this loan level
data must cover homeowner behavior over an extended time period. As is discussed more fully
· below, the analysis presented in this paper is based on a unique data set that meets all of these
criteria.
The primary empirical findings of this analysis can be summarized as follows. First, the
results reconfirm the importance of individual borrower and property characteristics in
prepayment behavior. Second, controlling for interest rate levels and volatility, points and fees,
and homeowners' equity and credit histories, the analysis strongly supports the hypothesis that
changes on both the supply and demand sides of the primary mortgage market have made
3
homeowners more inclined to refinance in the 1990s than was the case in the 1980s. Finally, the
analysis also finds that homeowners delay refinancing as interest rate volatility increases,
consistent with the conclusion of option valuation theory that the value of the call option
imbedded in the standard mortgage contract rises with volatility.
The plan of the paper is as follows. Section 2 presents the theory of the optimal refinance
decision rule. Section 3 describes the data set used in this analysis. Section 4 presents our
model specification and defines the explanatory variables. Section 5 presents the empirical
results of the model estimation. Finally, Section 6 concludes and presents policy implications of
this research.
2.
THE THEORY OF OPTIMAL PREPAYMENT
Interest Rates and Refinancing
The starting point for modeling prepayment behavior is the option pricing model (see, for
example, Follain, Scott, and Yang (1992)). The simplified premise is that prepayment is optimal
if the present value of an existing mortgage liability exceeds the present value of a replacement
loan by at least the total of all transaction costs associated with obtaining the new mortgage.
Suppose that borrower (I) takes out a mortgage loan at time t0;and that the expected terminal date
of the mortgage is T1., which is equal to or less than the maturity of the loan.
Let P(T.,t,r .)
I
Cl
represent the present value at month t (t=to;,·•·,T;) of the stream of payments based on the original
contract interest rate (rc;) of the I-th household's (callable) mortgage discounted at currently
prevailing interest rates (r,.,;) . Similarly, P(T;,t,r,.,;) is the present value of the stream of
payments on the same dollar amount of indebtedness based on prevailing market interest rates
4
(rm,;) and discounted at that same rate, i.e. the book value of the loan. A household seeking to
minimize the present value of its mortgage financing cost will prepay if
(1)
where (TC") equals the sum of points, fees and all other costs of making the transaction.
Volatility and the Option Replacement Cost
Expression ( 1) ignores the effect of uncertainty about future levels of interest rates on the
refinancing decision. Of course, interest rates are volatile; the more volatile rates are expected to
be in the future the less likely one is to exercise the embedded call option today since rates may
decline further in the future. Put differently, higher expected future volatility increases the value
of the call option. However, that volatifity increases the value of an option "in-the-money"--the
option in the existing mortgage--much more than',111 option well out of the money--the option in
the replacement financing. Thus, the difference: in value between these two options is another
component of the costs of refinancing faced by the homeowner.
To capture this effect, the value of a callable mortgage asset can be expressed as the value·
of a noncallable bond less the value of the imbedded call option. Abstracting from the subscript
I,
P(T,t,r1) = B(T,t,r1)
-
V(T,t,a,),
where B(•) is the annuity value of the stream of monthly mortgage prepayments and V(T,t,a,)
5
market conditions, transaction costs (TC,;) reflect a mix of market and individual factors.
Conceptually, transaction costs can be divided into a number of distinct components: (I) direct,
out-of-pocket expenses associated with prepaying the existing loan and obtaining replacement
financing (e.g., points and fees, prepayment penalties, and legal expenses) (TCPOINTS); (2)
additional out-of-pocket expenses, such as higher points and/or interest rate, and additional
documentation required because of a poor credit rating or score (TCcREDrr); (3) costs such as
mortgage insurance that may result from low equity in the property (TCLTV); and finally (4)
frictional costs that may reflect the homeowner' s time lost, the length of the application process,
and the unsophistication of the borrower (TCFRICTION). As shown in Figure 3, the first
component of transaction costs--points and fees--have fallen over time, from 2.5 percent of the
Joan value in 1983 to around I percent at the end of 1995, likely reflecting both better technology
and increased competition. Transactions costs associated with poor credit ratings may also have
been reduced by innovations such as credit scoring and subprime lending, which provide lenders
with a more efficient basis for pricing credit risk. Less well measured will be the cost of
searching for and comparing different lenders or the burden of completing applications, most of
which should have been moderated by more open competition and technological advances.
Our analysis assumes that transaction costs are reflected in the ap11lication fees, points,
mortgage insurance premiums, and other charges levied at the time of Joan application or
origination or are amortized in the form of a higher interest rate charged on the Joan itself. Either
way, total transaction costs are likely higher for credit- and/or collateral-constrained borrowers.
Note also that, to the extent transaction costs have important fixed components, they may not rise
proportionally with Joan size, causing refinancing behavior to differ accordingly.
7
denotes the value of the embedded call option at period t. 5 The value of this option, often
referred to as the "time value" of the mortgage, depends on the expected holding period of the
mortgage (T-t) and the volatility of the noncallable asset (0 1). As a result, we can replace (I)
with
(3)
where
(4)
The left-hand-side of equation (3) represents the "intrinsic value" of refinancing or the financial
gain from refinancing at the currently prevailing market interest rate. The variable u(T;,t,01),
which we label the option replacement cost, enters the analysis as the mechanism whereby
interest rate volatility affects refinancing behavior. Gilberto and Thibodeau ( 1989) provide
evidence that increased interest rate volatility reduces refinancings. This finding suggests that
the effect of volatility on the value of the call option imbedded in the existing mortgage exceeds
the effect on the value of the option imbedded in a replacement loan. This in tum is consistent
with the prediction of option theory that the effect of volatility on the value of an option is
greatest when it is in-the-money (see Hull (1993), Section 13.9).
Transaction Costs
While the intrinsic value of a loan and the option replacement cost are based on strictly
1'he price of an option also depends on the value of the capitalization factor measured indirectly by the
risk-free interest rate. For simplicity, however, we assume that this risk-free rate is constant over time.
6
3.
DATA
The data for this study were obtained through the Mortgage Research Group (MRG) of
Jersey City, New Jersey. Until late 1996, MRG maintained a data base on roughly 42 million
residential properties located in thirty-six states. In addition to information pertaining to
the
original purchase of a property, such as date of closing, purchase price, original mortgage
loan
balance, and maturity and type of mortgage, data on subsequent refinancings, sales, and,
is some
cases, defaults, were also included. 6 In addition to the property and loan characteristics,
the
database also contains snapshots of the credit histories of the occupants of the properties,
derived
from TRW Information Services.
Aside from limiting the sample to complete observations, we further restricted it to a
manageable size for computational purposes. Fist, we selected four clusters of counties in
different regions of the country. 7 Next, we identified for each property the most recent purchas
e
transaction, going in some cases back far as January 1984. The mortgages on many of these
properties were subsequently refinanced, while the others had no further transactions recorded
through the end of our sample period, December 1994, creating a zero-one, norefinance/refinance observation. We also limited our sample to fixed-rate mortgages outstand
ing
"The primary sources of this informatio.n are the records of county recorders and tax assessors.
7
1n the East, we chose four counties surrounding New York City (Orange County in New York State,
and
Essex, Bergen, and Monmouth Counties in New Jersey). In the South, we chose six counties in central
Florida
(Citrus, Clay, Escambia, Hernando, Manatee, and Marion). In the Midwest, we chose Cook County
and five
surrounding counties in Illinois (Dekalb, DuPage, Kane, McHenry, and Ogle). In the West, we selected
Los
Angeles, Ventura, and Riverside Counties in California. Selection ofthese diverse areas increases
our confidence
that our findings are general, particularly since the behavior of home prices in these four regions has
been rather
different.
8
for a year or more, leaving the more complex decision to refinance alternative mortgage types for
further study. The resulting sample consists of 12,835 observations, of which slightly under onethird were refinanced. 8 The credit snapshots attached to these propertynoan observations are as
of the second quarter of 1995.
4.
MODEL SPECIFICATION AND VARIABLE DEFINmONS
An Econometric Model of Prepayments
Several researchers have reported empirical models predicting prepayments from
property-level observations. For example, using a hazard model of mortgage terminations,
Caplin, Freeman, and Tracy (1997) provide evidence that homeowners with shrunken home
equity are Jess likely to prepay. Using Jogit models, Cunningham and Capone (1990) and Archer,
Ling, and McGill ( 1995) also find importance effects of home equity on the probability of
refinancing.
Our study uses Cox's proportional hazard framework to estimate a model of monthly
prepayments. The implied dependent variable in hazard analysis is the duration of time until the ·
Joan is prepaid (or, inversely, the conditional monthly probability of refinancing). The
proportional hazard model is given by
(5)
where ("t) denotes duration of the mortgage Joan and the vector xri_includes all explanatory
8
For multiple refinancings, we considered just the first one. In addition, we excluded from the sample loans
that subsequently defaulted.
9
variables. In the framework developed above, the duration of the i-th homeowner at month t is
given by , =t-to;· The function ho(,), the "baseline hazard function", equals the hazard for a
0
household whose exogenous vector is zero.
The parameter vector
allows us to estimate the
f3
f3
is estimated using partial maximum likelihood (PML), which
coefficients in the proportional hazard model without specifying a
functional form for the baseline hazard. The PML estimator is consistent, has an asymptotically
normal distribution, and has been found to have asymptotic relative efficiency (see Efron (1977)).
Variable Definitio ns
The endogenous variable in the proportional hazard model is the duration until the time
of refinancing measured in months. The vector of explanatory variables x,;. controls for the
intrinsic value of refinancing, the option replacement cost, and transaction costs.
We measure the intrinsic value of refinancing by the present value annuity ratio proposed
by Richard and Roll (1989):
PVALUE,;
(6)
where rci · again represents the coupon rate on the existing loan of the I-th borrower and r,,,,; is
the current market rate or rate at which that borrower could refinance. In practice, borrowers
may select from a menu of rate and point options, paying points in exchange for a lower coupon
rate, as well as choosing from differing maturities. This creates a difficulty in comparin g the
coupon rate on the existing loan with the currently prevailing market rate. To deal with this, we
standardized the rates by assigning re; as the average Freddie Mac commitment rate on a 30-year,
10
fixed-rate mortgages for the month that loan was closed. 9 This rate is for so-called A credits, or
borrowers who could meet the Fannie Mae and Freddie Mac underwriting criteria. Note that this
original rate is fixed for the life of the loan while the prevailing market rate on newly issued
loans, r mri' varies monthly. In computing PVALUE,;, we assume that the expected holding
period of the loan is 30 years. We reestimated the model using 15- and 20-year horizons and
found that the proportional hazard estimates arc: quite robust to the choice of assumption about
maturity.
As noted, the option replacement cost will vary positively with expected volatility.
However, in contrast to the intrinsic value of refinancing, which can be calculated directly, the
option replacement cost is unobservable. But we can observe the standard deviation of the price
of the noncallable asset. Thus, to estimate the effect of volatility on the decision to refinance, we
used the implied volatility from options on 10-ye'l\1" U.S. Treasury note futures contracts
(VOLATILITY). 10
The model controls for three of the four types of transaction costs or frictions discussed
above. Points and fees (TCPOINTS) comprise an important fixed cost which vary not only with
mortgage market conditions but also reflect individual borrowers' menu choice. Accordingly, we
included in the set of explanatory varial?les the average points and fees on mortgages issued
expressed as a percent of the loan amount (POJNTS).
9
The Freddie Mac mortgage interest rate series is published weekly. The interest rate is a contract rate with
associated points and fees also published. By design, the points and fees are reasonably.stable over time ..
1
°rhe implied volatility data are from the rolling 3-month futures options contracts traded on the Chicago
Board of Trade. We also experimented with several statistical variance computations or econometric projections of
actual rate volatility. These alternative measures of volatility yielded qualitatively similar results. In this study we
present our findings with respect to the implied volatility because this is conceptually preferable.
11
Borrowers also may face additional frictions because of poor credit history (TCCREDIT).
To capture this effect, we use the worst-ever credit rating from the borrower 's credit snapshot
(WRSTEVER). This credit rating represents the worst payment experience across all credit lines
over the individuals entire credit history. It is expressed as the number of days late; for example,
a worst-ever rating of 90 means that at one time the individual was reported ninety days late on a
credit card, car loan, mortgage, or other debt. The best possible worst-ever rating is a I, meaning
no late payments ever. On the other end of the scale, a worst-ever score of 400 means that a
lender has charged off a debt of that borrower (see Peristiani et al (1997) for additional details
on the credit snapshot)."
Another potential transaction friction stems from the amount of equity a borrower has in
the property (TCLTV). Borrowers applying for a mortgage loan that has a loan-to-value ratio
greater than 80 percent are usually required to take out private mortgage insurance, which
typically involves some payment at closing as well as a higher interest rate on the loan. If the
value of a property has fallen, in which case the loan-to-value ratio may exceed 100 percent, the
borrower would most likely be unable to refinance. We measure the effect of home equity by the.
ratio of the outstanding mortgage loan balance to the current value of the property (LTV). The
current value of the property is the original purchase price adjusted for local home price
movements. 12
11
An alternative credit measure ~uld have been the worst now credit rating or the worst payment
experience as of the date of the credit snapshot. Earlier experimentation found that the effect of a bad worst ever
rating on refinancing probabilities lingers even after the worst now has improved relative to worst ever. For this
reason, we chose the worst ever rating.
12
Current home prices were estimated by adjusting the purchase price for movements in county-level home
price indexes from Case, Shiller, Weiss Inc. Outstanding loan balance was inferred from the original loan amount,
the contract interest rate, and the original maturity.
12
While our analysis controls for many of the frictions associated. with refinancing, it is
impossible to fully account for all such frictions since they may depend on the efficiency of
mortgage lenders and the level of sophistication of borrowers. A homeowner' s decision about
whether or not to refinance a loan may also depend on the loan size involved. To the extent
certain costs of refinancings are fixed rather than proportional to the loan size, larger loans may
be refinanced more readily. Thus we include a variable measuring the size of the monthly
payment (SIZE), with the expectation that it should be positively correlated with refinancings,
other factors equal.
Table I summarizes the explanatory variables used in the estimation, with separate means
shown depending on whether or not the loan was refinanced.
5.
STATISTICAL RESULTS
Table 2 shows the partial maximum likelihood estimates of the parameter vector
13.
The
Waid chi-square statistics presented at the bottom of the table reject the null hypothesis that
H 0:f3=0. The first column in the table presents the coefficient estimates for the entire panel.
We find that the coefficient on PVALUE is positive and significant, as predicted. VOLATILITY
has a significant negative effect on the refinancing decision, consistent w.ith the hypothesized
effect of interest rate uncertainty on the option replacement cost. Also as expected, POINTS
have a significant negative effect..The negative and significant coefficient on LTV confirms that
equity-constrained borrowers are less likely to refinance. Similarly, a high WRSTEVER score
reduces the refinancing probability, although this effect is quantitatively less pronounced·than
that of LTV. The small effect of credit quality may reflect measurement problems with the
13
worst-ever credit snapshot; if available, a continuously evolving credit measures might produce a
stronger estimated effect on refinancing. 13 Finally, the size of the monthly payment (SIZE) has
the predicted sign and is statistically significant, suggesting that the size of the loan may provide
some incentive to refinance beyond that reflected in PVALUE and that there may be important
fixed costs associated with refinancing.
Credit and Collateral Subgroups
Since earlier research has found that credit and collateral variables can interact with the
other explanatory variables, we estimated the model separately for different credit and collateral
subgroups. The findings, shown in Table 2, Columns 2 through 5, are again consistent with the
notion that credit and equity affect refinancing probability, but a channel of effect is clarified:
The estimated sensitivity of refinancing probability to PVALUE is appreciably lower for creditand equity-constrained borrowers. This is consistent with the separate findings of interaction
effects between refinancing incentives and home equity (Caplin, Freeman, and Tracy (1997)) and
between refinancing incentives and credit ratings (Peristiani et al (1997)).
The impact of a poor credit rating can be illustrated by estimating the survival function of ·
loans, that is, the cumulative likelihood of a loan "surviving" (i.e., not being refinanced) over
time, for borrowers in different WRSTEVER categories. In the partial likelihood framework, the
survival function can be estimated from
13
For example, the delinquency may have occurred some time ago, been on a less significant credit line
(for instance, a store card rather than a mortgage), or the borrower may have been able to provide a reasonable
explanation for the delinquency. Also, Peristiani et al (1997) report evidence that an improvement in credit history
(worst now better than worst ever) increases refinancing probabilities, but not enough to completely erase the effect
of a poor worst-ever score.
14
(7)
The variable So(,)represents the baseline survival function. An estimate of the baseline function
is obtained using a nonparametric maximum likelihood method. Figure 4 presents survival
functions for two WRSTEVER categories, good credits (WRSTEVER= I; the bottom line) and
poor credits (WRSTEVER=400; the top line). These estimated survival functions indicate that-under the market conditions faced by the borrowers in our sample--nearly 13 percent of the good
credits had refinanced after I 00 months (eight years and 4 months) versus just about 3 percent of
the poor credits. 14 This underlines the need to properly control for these factors, which enter the
equations in nonlinear ways, in comparing refinancing propensities over time.
Comparing Refinancing Propensities in the 1980s and 1990s
To explore the possibility that refinancing behavior has changed, we divided the sample
between mortgagors that purchased their homes during 1984-90 and those who purchased during
1991-94. Because the sample includes just one credit snapshot ( 1995Q2), we were hesitant to
assign_that credit information to refinancing behavior over the period of a decade. Nevertheless,·
based on the Table 2 results, credit is demonstrably important and should not be ignored in
intertemporal comparisons. Therefore, in this sample splitting exercise we controlled for credit
rating by limiting our sample to good credits (WRSTEVER=l). By taking this approach, we
arguably reduce the generality of our results somewhat, since an increase in the incidence of
weak credit ratings in the 1990s would be at least a partial offset to any increased propensity to
14
It is important to remember that refinancing probabilities are considerably smaller than overall
prepayment probabilities, which include sales and defaults. Thus, the survival rates in this sample (which excludes
sales and defaults) will appear correspondingly higher.
15
refinance among good credits. On the other hand, limiting the sample to good credits makes ours
a purer test of the hypothesis that structural changes have increased refinancing probabilities.
The proportional hazard model, excluding the credit rating variable, is estimated for the
two subsamples to test the null hypothesis H :
0
1384 _90 =13 91 _94 , where the subscripts identify the
date range of home purchases in the respective subsamples. The coefficient estimates for
1391 _94
are shown in column 3 of Table 3. Estimating the model for borrowers that took out a loan
during 1984-90 is not completely straightforward because in this case the sample "spills over"
into the 1990s (that is, purchasers in the 1980s are still at risk of refinancing their purchased
mortgages in the latter decade). To address this potential problem, the proportional hazard
coefficients presented in the column 1 of Table 3 represent early period purchases ( 1984-90) but
with the data truncated after 1990 -- focusing the estimation more closely on refinancing
behavior during 1984-90. For comparison, column 2 shows the coefficient estimates for
borrowers in 1984-90, but continuing the sample into the 1990s.
The differences in the estimated coefficients for the subperiods are striking. Households
that purchased during 1991-94 (column 3) are much more responsive to the intrinsic value of
refinancing (PVALUE). At the same time, the estimated coefficient of POINTS has the
predicted sign and is highly statistically significant in both subperiods. The coefficient on
VOLATILITY is not statistically significant in the earlier subperiod while it is highly significant
with predicted sign in the latter period. The coefficient on LTV does not have the predicted sign
in the earlier period, perhaps because home prices were generally rising quite rapidly over that
period.
The difference in sensitivity to PVALUE in the two subperiods is quite surprising given
16
that, on average, the values of PVALUE that mortgagors were exposed to over the two
subperiods were roughly comparable. 15 This is quite strong support for our hypothesis of a
structural change and is consistent with the anecdotal conclusion that the interest rate differential
needed to induce a mortgagor to refinance has declined. Also supportive of this hypothesis is the
fact that the size of the coefficient on the variable SIZE declines very sharply in the latter period,
consistent with the idea that fixed-cost transaction frictions which have declined over time. In
short, the results are quite consistent with the idea that lower transactions costs (measurable and
otherwise), and perhaps increased sophistication of borrowers, have increased the propensity to
refinance. As the Waid test shows (bottom of Table 3}, borrowers in the 1990s continue to
exhibit a much greater willingness to refinance.
Figure 5 contrasts the mortgage loan survival experience of the 1990s with that of the
1980s. For the 1980s we simulate the survival function two ways. The top curve represents
'
1984 to 1990 parameter values (column I of Table 3) with values of the explanatory variables for
the same period.
16
The middle curve represents 1984-90 parameter values but with values for the
explanatory variables from the 1991-94 period. The distance between the top and middle
survival functions reflects the effects of the differing exogenous variables (including,
importantly, about a 50 basis point diff~rence in average points and fees) between the two
periods. 17
15
In fact, mortgage holders enjoyed more favorable interest rate spreads during 1984-1990. The average
interest spread during the period 1991-94 was around 60 basis points compared to 125 basis points during 1984-90.
~
.
.
The survival function for the untruncated 1980s sample (column 2 of Table 3) is very similar.
17
.
The average coupon rate spread (rm,;-rci) in the period 1984-90 was about 150 basis points. By contrast,
the mean coupon spread during 1991-94 was roughly 60 basis points, hence not contributing to higher refinancing
rates in the 1990s subperiod.
17
The bottom survival curve represents 1991-94 parameter estimates and 1991-94 values of
the explanatory variables. Hence the middle and bottom survival curves compare individuals
exposed to the same explanatory variables, but with different responses to those conditions as
represented by the differences in the estimated coefficients. The distance between these two
curves represents the difference in refinancing behavior that can be attributed to structural change
in the mortgage market, above and beyond the changes in measurable transaction frictions such
as points and fees. For example, after four years in the I 990's, nearly 14 percent of the purchase
mortgages loans had been refinanced. In contrast, under I 980s behavioral response, cumulative
refinancings over the first four years would have totaled only 9 percent.
6.
DISCUSSION AND CONCLUSIONS
We developed an empirical model to test whether structural changes in the U.S. mortgage
market have affected mortgagors' refinancing behavior. We hypothesized that the intrinsic
benefit required to trigger a refinancing has become smaller, due to a combination of
technological, regulatory, and structural changes that have made mortgage origination more
competitive and more efficient. To test this hypothesis, we estimated an empirical hazard model
of loan survival for two time subperiods, using a database that allowed us to carefully control for
homeowners' credit ratings, equity, loan size, and measurable transaction costs. Overall, we are
confident that our hypothesis has been tested on the basis of a reasonably comprehensive model
of individual and market determinants of refinancings.
Our findings strongly confirm earlier findings that credit ratings and home equity have
significant effects on refinancing probability. In addition, we provide evidence that homeowners
18
postpone refinancing in the face of increased interest rate volatility, consistent with option value
theocy. Finally, our results clearly support the hypothesis that structural change in the mortgage
market has increased homeowners' propensity to refinancing. This conclusion emerges from two
findings. One is that measurable transaction costs, such as points and fees and other fixed costs,
are quite important in the refinancing decision and that those costs have declined significantly in
the 1990s relative to 1980s--a development we attribute to increased efficiency and competition
in mortgage origination. Secondly, even after controlling for points and fees, loan size, and other
important variables, refinancing probabilities were considerably higher in the latter period. This
we attribute to declines in nonmeasurable frictions, which likely takes the form of aggressive
solicitations of refinancings by lenders, which have the effect of disseminating information faster
and more broadly, as well as increased financial sophistication among homeowners.
19
REFERENCES
Archer, Wayne, David Ling, and Gary McGill. "The Effect of Income and Collateral Constraints
on Residential Mortgage Terminations," NBER Working Paper No. 5180, July 1995.
Bradley, Michael G., Stuart A. Gabriel, and Mark E. Wohar. "The Thrift Crisis, MortgageCredit Intermediation, and Housing Activity," Journal of Money, Credit, and Banking 27
(1995), 476-497.
Caplin, Andrew, Charles Freeman, and Joseph Tracy. "Collateral Damage: How Refinancing
Constraints Exacerbate Regional Recessions," Journal of Money, Credit, and Banking
(forthcoming 1997).
Cunningham, Donald, and Charles A. Capone, Jr. "The Relative Termination Experience of
Adjustable to Fixed-Rate Mortgages," Journal of Finance 45 (December 1990).
Efron, B. "The Efficiency of Cox's Likelihood Function for Censored Data," Journal of the
American Statistical Association 72 (1977), 557-565.
Follain, James R., James 0. Scott, TL Tyler Yang. "Microfoundations of a Mortgage Prepayment
Function," Journal of Real Estate and Economics 5 (1992), 197-217.
Glaesar, Edward R. and Hedi D. Kallal. "Thin Markets, Asymmetric Information, and Mortgage
Backed Securities," Journalof Financial Intermediation 6 (1997), 64-86.
Hendershott, Patric H. and Kevin Villani. The Evolution of the Housing Finance System.
American Enterprise Institute, 1978.
Hull, John C. Options, Futures, and Other Derivative Securities. Englewood Cliffs, New
Jersey: Prentice-Hall, 1993.
Richard, Scott F., and Richard Roll. "Prepayments on Fixed-Rate Mortgage-backed
20
Securities," The Journal of Portfolio Management 15 (Spring 1989), 73-82.
Peristiani, Stavros, Paul Bennett, Gordon Monsen, Richard Peach, and Jonathan Raif. "Effects of
Household Creditworthiness on Mortgage Refinancings," Research Paper #9622, August
1996.
Weicher, John C. "The New Structure of the Housing Finance System," Federal Reserve Bank of
St. Louis, July/August 1994.
21
TABLE 1
SUMMARY STATISTICS
Mean
Explanatory
Variable
Description
Refinancing
No
Refinancing
WRSTEVER
Worst delinquency ever (1 = good credit, 30, 60,
90, 120, 150, 180, 400=default)
64.79
99.71
PVALUE
Present value ratio as defined by equation (9)
(percent)
117
Ill
LTV
Current loan-to-value (percent).
54.94
66.49
VOLATILITY
Implied volatility on options on the 10-year
treasury note futures (basis points).
6.63
.7.06
POINTS
Initial fees and point changes on conventional
home mortgages. National average for all major
lenders (percent).
1.99
2.03
AGE
Age of loan (measured in months).
48.17
53.59
SIZE
Logarithm of original loan balance (balance
measured in thousands of dollars).
11.65
11.39
4226
497243
Number of monthly
observations
TABLE2
FACTORS INFLUENCING THE DECISION TO REFINANCE: PROPORTIONAL HAZARD MODEL
Explanatory Variable
DUM-NY
DUM-IL
DUM-FL
WRSTEVER
PVALUE
VOLATILITY
LTV
POINTS
SIZE
All
Borrowers
0.688***
(157.97)
-0.947***
(194.96)
0.959***
(292.70)
-0.001***
(65.39)
1.210***
(40.07)
-0.454***
(371.26)
-0.019***
(763.55)
-5.350***
(2462)
0.326***
(130.66)
Good Credit
(WRSTEVER=l)
Poor Credit
(WRSTEVER::400)
LTV,;;80
LTV>80
LTV,;;80
LTV>80
0.966***
(130.41)
-0.435***
(21.52)
1.076***
(152.97)
0.511***
(7.85)
0.258
(1.23)
1.307***
(45.10)
0.747***
(12.49)
-1.81 ***
(56.97)
1.175***
(37.55)
0.112
(0.12)
-1.862***
(12.67)
1.003***
(8.05)
1.403***
(22.70)
-0.455***
(161.53)
-0.021***
(327.45)
-5.836***
(1261)
0.271***
(43.27)
1.463*''·*
(6.84)
-0.532***
(75.50)
0.010*
(2.65)
-3.915***
(165.25)
0.740***
(71.07)
-2.388***
(13.18)
-0.317***
(13.53)
-0.037***
(163.16)
-4.91 ***
(179.49)
0.146
(2.01)
-1.957**
(4.32)
-0.651 ***
(35.65)
0.015
(2.00)
-3.166***
(37.61)
0.788***
(20.30)
155.43
730.45
Wald chi-square
6284.80
2987.14
379.55
50339
54369
497243
166484
83552
Censored obs.
205
332
4226
1864
637
Refinancing obs.
NOTES: The symbols(***), (**), and(*) indicate statistical significance at the 1-, 5-, and IO-percent level,
respectively. Table l describes in more detail the explanatory variables.
TABLE3
THE Wil.LlNGNESS TO REFINANCE OURJNG THE 1980s AND 1990s
(Numbers in parentheses represent Wald chi-square statistics).
Purchases in 1984 -1990
Purchases in 1991-1994
Variable
1984-19908
1984-1994b
OUM-NY
-17.91
(0.00)
1.448***
(261.38)
-0.849***
(47.12)
OUM-IL
-2.129***
(56.06)
-0.777***
(51.17)
0.060
(0.18)
OUM-FL
-0.436**
(5.02)
1.388***
(252.50)
-0.441 ***
(10.05)
PVALUE
-1.79*
(2.34)
0.832***
(6.86)
3.76***
(44.23)
LTV
0.033***
(41.88)
-0.008***
(43.38)
-0.031***
(544.91)
VOLATILITY
0.087
(1.23)
-0.391 ***
(101.29)
-0.623***
(141.43)
POJNTS
-3.430***
(48.87)
-3.89***
(405.78)
-5.118***
(305.05)
SIZE
0.536***
(24.03)
0.413***
(76.81)
0.052***
(0.75)
2
235.80**.*
1686.68***
1317.95***
2
X test
Ho:P9,-94 =P84-90
297.85***
319.74***
Refinancing obs.
277
1486
1015
Censored obs.
97520
160095
89941
X testHo:P=0
8
Sample of homeowners is truncated after 1990.
t>rroportional hazard estimates for the complete spell (e.g., last observation of panel ends in
December 1994 or at the month of refinancing, which ever comes first).
NOTES: The symbols(***),(**), and(*) indicate statistical significance at the 1-, 5-, and 10percent level, respectively. Table 1 describes in more detail the explanatory variables. x2 values
are Wald statistics.
Figure I: Prepayment Speeds on FNMA MBS Backed by 30-year Fixed Rate Mortgages
Single Monthly Mortality
8
8
Unseasoned Moderately Seasoned Seasoned
0-30 months
31-60 months
60+ months
A
6
"
I \A
I'
j
\I\
I ft
4 •-
A /I
~\
1
I
n ,,:~,'!'\
! i
I
I :ii
ii
/:\\/: :H:g
,.
\
/\
'-
'•
\i
\i
\
\
\
,,
/,;,."..._
I'
I
o.:
I
l\/1
: 1 .
,,
1
1
1111
t
;
I I
=,
I I
1
1 \ I
,
If
/\
I ' '1 -
i \_.
0
f
\\
/\
.....
i ! ~
I,\ /:~~\/:. : \
I
:-
i
\
\
/IV,'
H" . . ,
~.:,..""
11 ~;\1i
1
•\
2
6
\
\
\
•....
, .. ,__ f ...":'- \ / - , \
..,_,.... ·...\,,.. ............
f
/t
II
If
,1
i
I
4
\1:\
u,i
, I:•
\\
11
I
1.
'I\I
"
ll
I
l\v1
\
I
\
:"••
I
,.
:,
,.._..'.,·l : . ·····
..,
...\
' ....
2
/\
I'
•\
I\',
'
\-.,\.I
•
•.
Feb-82
0
Aug-83
Feb-85
Aug-86
Feb-88
Aug-89
Source: Global Advanced Technology Corporation
Feb-91
Aug-92
Feb-94
Aug-95
Feb-97
Figure 2: Primary Mortgage Market: Market Shares by Type of Lender
(Percent of 1-4 Family Originations)
Annual Percent of Total Dollar Volume of Loans
70 , - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - , 70
Thrifts
60
50
..
•
..
•••
•••• ••••
••••••••.
••
••
.,.
•• I
•••
I•
60
••
••
••
•
..•
••
.• :\.•
. .•\
I•
•····· ...•••
•
..•• ••.
-• 50
..•
•
.. ..••·····••
.
.......
40
Mortgage Bankers
#.
40
•••
·,c \
30
-- - - ~
'
--,
,.,,
'
~
/
s-i 30
I •••• \
•
.....
.•·····•... - - '
..~..
•• •-...
'
....
• •• - 1
20
10 L.J.___,;.-'--....L.-..J'--J._---'--L-...J....----1--'--....l..-'--..l...--1.-.l.-....J...__.J_ _.___.__,__....L---1.--'-----'--..J'----'-----'-'
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
lO
20
•••••
I
Commercial Banks
Source: U.S. Department of Housing and Urban Development, Survey of Mortgage Lending Activity
Figure 3: Initial Fees and Charges on Conventional Loans Closed
Percentage of loan amount
3.0 , - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - ,
2.5
2.0
1.5
1.0
0.5 I 1,,,,,,,,,,,1,,,,,,,,,1,1,,,,,,,,,,11,,,,,,,1,,1!111111,,,,,1,,,,,,,,,,,1,,,,,,,,111!11111111111!1111,111111!1111111,11111,11111111111,,,,,,,,,,1,,,,,,,111111111111111, I
Jan-83 Jan-84 Jan-85 Jan-86 Jan-87 Jan-88 Jan-89 Jan-90 Jan-91 Jan-92 Jan-93 Jan-94 Jan-95 Jan-96
Figure 4. Survival Function by Credit Level
1. 00
=··-----------------------------·-----------
0.99
-....._
·-----------...
0.98
----,.......,
,.
''
0.97
0.96
s
0.95
R o.94
V
I 0.93
V
A 0.92
L
0. 91
0.90
0.89
0.88
0.87
0.86
.,.
0
10
20
30
40
50
60
70
80
90
100
AGE (in months)
I
CREDIT:
WRSTEVER=l
-------· WRSTEVER=400
I
110
Figure 5. Survival Function Before and After 1990
~-------
1.00
---------------·----·
0.99
0. 9 8
0.97
0.96
0. 9 5
s
R o.94
V
I 0.93
V
A O. 9 2
L
0. 91
0.90
0. 8 9
0. 8 8
0.87
0.86
T
0
10
20
30
40
50
60
AGE(in months)
PERIOD:
1991-94: X(90s)
1984-90: X(90s)
-------· 19 8 4 - 9 0 : X ( 8 0 a)
FEDERAL RESERVE BANK OF NEW YORK
RESEARCH PAPERS
1997
The following papers were written by economists at the Federal Reserve Bank of
New York either alone or in collaboration with outside economists. Single copies of up
to six papers are available upon request from the Public Information Department,
Federal Reserve Bank of New York, 33 Liberty Street, New York, NY 10045-0001
(212) 720-6134.
9701. Chakravarty, Sugato, and Asani Sarkar. "Traders' Broker Choice, Market Liquidity, and
Market Structure." January 1997.
9702. Park, Sangkyun. "Option Value of Credit Lines as an Explanation of High Credit Card
Rates." February 1997.
9703. Antz.oulatos, Angelos. "On the Determinants and Resilience of Bond Flows to LDCs,
1990 - 1995: Evidence from Argentina, Brazil, and Mexico." February 1997.
9704. Higgins, Matthew, and Carol Osler. "Asset Market Hangovers and Economic Growth."
February 1997.
9705. Chakravarty, Sugato, and Asani Sarkar. "Can Competition between Brokers Mitigate
Agency Conflicts with Their Customers?" February 1997.
9706. Fleming, Michael, and Eli Remolona. "What Moves the Bond Market?" February 1997.
9707. Laubach, Thomas, and Adam Posen. "Disciplined Discretion: The German and Swiss
Monetary Targeting Frameworks in Operation." March 1997.
9708. Bram, Jason, and Sydney Ludvigson. "Does Consumer Confidence Forecast Household
Expenditure: A Sentiment Index Horse Race." March 1997.
9709. Demsetz, Rebecca, Marc Saidenberg, and Philip Strahan. "Agency Problems and Risk
Taking at Banks." March 1997.
.
9710. Lopez, Jose. "Regulatory Evaluation of Value-at-Risk Models." March 1997.
9711. Cantor, Richard, Frank Packer, and Kevin Cole. "Split Ratings and the Pricing of Credit
Risk." March 1997.
9712. Ludvigson, Sydney, and Christina Paxson. "Approximation Bias in Linearized Euler
Equations." March 1997.
9713. Chakravarty, Sugato, Sarkar, Asani, and Lifan Wu. "Estimating the Adverse Selection
Cost in Markets with Multiple Informed Traders." April 1997.
9714. Laubach, Thomas, and Adam Posen. "Some Comparative Evidence on the Effectiveness
oflnflation Targeting." April 1997.
9715. Chakravarty, Sugato, and Asani Sarkar. "A General Model of Brokers' Trading, with
Applications to Order Flow Internalization, Insider Trading and Off-Exchange Block
Sales." May 1997.
9716. Estrella, Arturo. "A New Measure of Fit for Equations with Dichotomous Dependent
Variables." May 1997.
9717. Estrella, Arturo. "Why Do Interest Rates Predict Macro Outcomes? A Unified Theory of
Inflation, Output, Interest and Policy." May 1997.
9718. Ishii Jun, and Kei-Mu Yi. "The Growth of World Trade." May 1997.
9719. Kambhu, John. "Interest Rate Options Dealers' Hedging in the US Dollar Fixed Income
Market." May 1997.
9720. Kroszner, Randall, and Philip Strahan. "The Political Economy of Deregulation
Evidence from the Relaxation of Bank Branching Restrictions in the United States."
June 1997.
9721. Locke, Peter, Sarkar, Asani, and Lifan Wu. "Market Liquidity and Trader Welfare in
Multiple Dealer Markets: Evidence from Dual Trading Restrictions." July 1997.
9722. Zakrajsek, Egon. "Retail Inventories, Internal Finance, and Aggregate Fluctuations:
Evidence from U.S. Firm-Level Data." July 1997.
9723. Lown, Cara, and Robert Rich. "ls There An Inflation Puzzle?" August 1997.
9724. Chang, Angela, Chaudhuri, Shubham, and Jith Jayaratne. "Rational Herding and the
Spatial Clustering of Bank Branches: An Empirical Analysis." August 1997.
9725. Helwege, Jean, and Christopher Turner. "The Slope of the Credit Yield Curve for
Speculative-Grade Issuers." August I 997.
9726. Antzoulatos, Angelos. ''Non-Linear Consumption Dynamics." August 1997
9727. Connolly, Michelle. "Technology, Trade and Growth: Some Empirical Findings."
September 1997.
9728. Antzoulatos, Angelos. "Macroeconomic Forecasts under the Prism of Error-Correction
Models." September 1997.
9729. Black, Sandra "Do Better Schools Matter? Parental Valuation of Elementary
Education." September 1997.
9730. Walter, Christian, and Jose Lopez. "ls Implied Correlation Worth Calculating? Evidence
from Foreign Exchange Options and Historical Data." September 1997.
9731. Jayaratne, Jith, and Don Morgan. "Information Problems and Deposit Constraints at
Banks." October 1997.
9732. Steindel, Charles. "Measuring Economic Activity and Economic Welfare: What Are We
Missing?" October 1997.
9733. Campbell, John, and Sydney Ludvigson. "Elasticities of Substitution in Real Business
Cycle Models with Home Production." October 1997.
9734. Harrigan, James. "Cross-CountrY Comparisons of Industry Total Factor Productivity:
Theory and Evidence." November 1997.
9735. McConnell, Meg, and Gabriel Perez Quiros. "Output Fluctuations in the United States:
What Has Changes Since the Early 1980s?" November 1997.
To obtain more information about the Bank's Research Papers series and other
publications and papers, visit our site on the World Wide Web (http://www.ny.frb.org). From
the research publications page, you can view abstracts for Research Papers and StaffReports and
order the full-length, hard copy versions of them electronically. Interested readers can also view,
,download, and print any edition in the Current Issues in Economics and Finance series, as well
as articles from the Economic Policy Review.
@FedFRASER