View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

clevelandfed.org/research/workpaper/index.cfm

Working Paper 94 19

POSTED RATES AS SIGNALS
IN MORTGAGE LENDING MARKETS
by Robert B. Avery, Patricia E. Beeson, and
Mark S. Sniderman

Robert B. Avery is a professor in the Department of Consumer
Economics and Housing at Cornell University, and Patricia E.
Beeson is an associate professor of economics at the University
of Pittsburgh; both are research associates at the Federal Reserve
Bank of Cleveland. Mark S. Sniderman is senior vice president
and director of research at the Federal Reserve Bank of Cleveland.
An earlier version of this paper was presented in March 1994 at the
Conference on Information and Screening in Real Estate Finance,
sponsored by the Federal Reserve Bank of Philadelphia. The authors
extend special thanks to Leonard Nakamura for many thoughts and
suggestions, and to Pamela Rice for research assistance.
Working papers of the Federal Reserve Bank of Cleveland are
preliminary materials circulated to stimulate discussion and
critical comment. The views stated herein are those of the
authors and not necessarily those of the Federal Reserve Bank
of Cleveland or of the Board of Governors of the Federal
Reserve System.

December 1994

clevelandfed.org/research/workpaper/index.cfm

Abstract

In many metropolitan areas throughout the United States, a group of mortgage lenders
active in the market post a set of mortgage lending terms each week in a local newspaper. Casual
inspection of these postings over time suggests that some lenders usually advertise low rates
relative to the market mean, while others tend to be above market. Furthermore, at any point in
time, the distribution of posted rates appears to vary considerably. Why do lenders post the rates
we see advertised, how frequently do they adjust the terms, and how does the market respond'?
In this paper, we discuss how lenders might use posted lending terms to signal 1) their
eagerness to take new loan applications and 2) their lending standards relative to other lenders in
their market. We demonstrate that lenders who lower their posted rates relative to their own
normal market position indeed attract more applicants. We also find that better quality applicants
are more likely to apply to low-rate lenders and that these lenders tend to sell off a larger portion
of the loans they originate, to apply less stringent underwriting standards, and to deny fewer loan
applications than do middle- or high-rate lenders. In our sample, the low-ratepow-risk lenders
tend to be independent mortgage banks or the mortgage subsidiaries of commercial banks and
thrifts. The high-ratehigh-risk lenders tend to be commercial banks and thrifts. These lenders
may be playing different roles in their respective markets.

clevelandfed.org/research/workpaper/index.cfm

I. INTRODUCTION

In many metropolitan areas throughout the United States, a group of mortgage lenders
active in the market post a set of mortgage lending terms each week in a local newspaper. The
rates are listed in a manner designed to permit an easy comparison of discount points and note
rates by credit shoppers in the market. Casual inspection of these postings over time suggests that
some lenders usually advertise low rates relative to the market mean, while others tend to be
above market. Furthermore, at any point in time, the distribution of posted rates appears to vary
considerably.
Why do lenders post the rates we see advertised, how frequently do they adjust the terms,
and how does the market respond? In addressing these questions, we view lenders as using
posted rates to signal their desired current position in the market to potential borrowers.
Specifically, we think that lenders are trying to accomplish two objectives with their posted rate
practices. First, lenders move their rates relative to the market mean in order to affect the flow of
mortgage loan applications they might receive. Their purpose is to attract or discourage
applicants based on adjustments they desire to make in their loan portfolios. Second, we
conjecture that lenders tend to specialize in evaluating loan applicants of different quality types,
and that they use posted rates to signal the quality type sought. This signaling activity would not

be necessary in an economy characterized by perfect information. However, it is likely that
several imperfections are manifest in actual markets. To cite just one example, if potential
borrowers do not know the underwriting standards of each lender, and if the search is costly for
both parties, then a set of posted rates may lead to more efficient matches.

clevelandfed.org/research/workpaper/index.cfm

To investigate these issues, we collected published monthly mortgage lending terms for 87
mortgage lenders active in Cleveland, Columbus, and Detroit during 1990, 1991, and 1992. The
lenders consist of commercial banks, savings and loans, mortgage subsidiaries of depository
financial institutions, and independent mortgage banks. A second data source, obtained from
lenders under provisions of the Home Mortgage Disclosure Act (HMDA), contains information
about mortgage loan applicants to these lenders during this same period. We use the HMDA data
to construct a measure of the applicant quality attracted by each lender for which we have rate
data.
The paper is organized as follows. In section D, we describe the mortgage lending terms
data, the applications data, and sample design. In section III, we report the distribution of lending
rates in the markets examined. Section IV considers the role of posted rates as switches used by
lenders to regulate their application flows, and section V investigates the role of posted rates as
signals of the type of loan quality a lender seeks. Our conclusions and suggestions for further
research are presented in section VI.

D. DATA
Mortgage Loan Application and Disposition Data
Information used to calculate mortgage loan application and disposition rates for
individual lenders in the three cities examined was drawn from data collected under the 1989
revisions to HMDA. Nearly all commercial banks, savings and loans, credit unions, and other
mortgage lending institutions that have an office in a Me~opolitanStatistical Area (MSA) are

clevelandfed.org/research/workpaper/index.cfm

required to report annually on each mortgage loan application received. Lenders must provide
such information as loan amount, census tract of the property, loan guarantee (conventional,
Federal Housing Administration [FHA], or Department of Veterans Affairs [VA]), loan
disposition, race and gender of the applicant, and applicant income.' This study utilizes home
purchase loan filings for the Cleveland, Columbus, and Detroit MSAs during 1990, 1991, and
1992.
Raw HMDA data provided direct information on the denial rate (percentage of loan
applications that were not approved by the lender), total number of loan applications, and
percentage of loan applications taken from minority (non-white) applicants for each lender. These
variables could be summed over applicants to provide a measure of the monthly activity of each
sample lender. Information on the quality of loan applicants and the lender's underwriting
standard, however, could not be gathered so straightforwardly. We derived these variables using
the predictions of a model estimated with the entire HMDA dataset, including lenders outside the
three MSAs. The model was developed as follows.
We assumed that each individual mortgage application's risk could be represented as a
function of the applicant's characteristics (such as race and income), neighborhood (census tract),
market (MSA), and lender. Moreover, we assumed that the probability of an application being
denied is linear in its risk. This implies that the probability of a random loan application being
denied is also linear, i.e.,

1

S e e Canner and Smith (1991,1992) for a comprehensive discussion of HMDA.

3

clevelandfed.org/research/workpaper/index.cfm

where DENIAL is one if the ith application using the Lth lender in the Mth MSA and Tth census
tract is denied, and zero otherwise. MSA, TRACT, and LENDER are dummy variables indicating
which MSA, census tract, and lender the application relates to, and e is a residual. AC is a vector
of application characteristics reported in the HMDA data, including, race, marital status,
occupancy, income, loan amount, income-to-loan ratio, federal loan guarantee (FHA or VA), and
month of the year the application was acted upon.'
HMDA data for home purchase loan originations were used to fit model (1) separately for
1990, 1991, and 1992.~Although the basic model form is linear, we used splines and interaction
terms to reflect potential nonlinearities. The actual estimation was done in two steps. In the first
step, equation (1) was estimated with the individual application characteristics (AC) and separate
intercepts for each lenderlcensus tract combination included as single-componentfixed effects. In
the second step, an iterative procedure, equivalent to regressing the fixed-effects intercepts
against MSA, census tract, and lender dummies, was used to identify the MS A, tract, and lender
effects. Separate lender effects were estimated for each MSA, in effect defining lenders operating
in multiple MSAs as multiple

lender^.^

Alternative specifications such as a logistic or probit model could have been employed: There is no
particular theoretical reason to choose among these forms. Thus, the practical dictates of a very
large sample led to the choice of the linear probability specification.
a Samples for each year included all home purchase loan applications for 1-4 family residential units
acted upon (accepted or denied) by the lenders. This included 1,984,688 applications in 1990,
2,087,470 in 1991, and 2,400,875 in 1992. In a small number of cases, some values for certain
variables had to be imputed because they were not reported by the lender.
By construction, the MSA effects were normalized to have overall sample means of zero; within
each MSA, lender and tract means were also normalized to zero. In cases where lender and tract
effects were not identified (a lender was the only lender in a tract and did all of its business there),
the effect was assigned to the tract.
4

clevelandfed.org/research/workpaper/index.cfm

We used the parameter estimates from equation (I), together with characteristics of the
applications received (AC, MSA, and TRACT), to predict denial rates on the basis of all factors

except lender.' This prediction, averaged over all applicants to a lender in a month, was used as a
measure of applicant quality. We estimated the lender's underwriting standard as the difference
between its denial rate for the month and the estimated quality of its applicants. If this residual
was positive, for example, it would indicate that the lender denied more applicants than would be
predicted that month, indicative of a tough underwriting standard. A negative residual would
indicate a looser standard.
Interest-Rate Data
Interest-rate data were collected from National Mortgage Weekly, a firm that telephones
lenders for pricing information which it then provides to newspapers to publish in their real estate
sections each week. The firm selects lenders that account for a large volume of loans in their
markets and asks for prices on a variety of mortgage loan products. National Mortgage Weekly
provided us with information on each of their weekly reportings for the 36 months spanning
1990-92 for the Cleveland, Columbus, and Detroit markets. We used the middle week of each
month as representative of the month and focused on one product, 30-year fixed-rate 10ans.~
Quoted interest rates were adjusted for lender points, with each point equaling a 114 percent
higher loan rate.

Parameter estimates for these regressions are not presented here because of space considerations.
They are available in Avery, Beeson, and Sniderman (1994a).
Clearly, this may be a biased estimate of a lender's true pricing position, since some lenders may
not make fixed-rate loans or may prefer to steer borrowers away from fixed-rate products.
Unfortunately, only the first-year rate was reported for variable-rate mortgages, whereas the
spread over the index would have been a more accurate measure of price. Since many firms offered
first-year "teasers" on variable-rate products, it was hard to compare the true prices of variable-rate
loans across lenders. For this reason, we decided to use only the fixed-rate price.
5

clevelandfed.org/research/workpaper/index.cfm

Study Sample
Unfortunately, the HMDA and National Mortgage Weekly samples did not mesh exactly.

In some cases, banks and savings and loans filed separate HMDA reports for their mortgage bank
subsidiaries, but may have posted only one rate for National Mortgage Weekly. In these cases,
the HMDA filings were consolidated. In other cases, lenders were not large enough to be
included in the price survey, but still filed HMDA reports. Somewhat inexplicably, price
information was sometimes reported for lenders who did not file HMDA reports (perhaps because
the market definition used by National Mortgage Weekly did not correspond exactly to the MSA
definition). In these cases, the lenders were not used. There were also a number of instances of
lenders being included in both data samples, but for only part of the study period. Since we were
interested in long-run behavior, we decided to eliminate all lenders that did not provide price
information for at least the first 24 months of the sample period or that didn't file HMDA reports
for at least 1990 and 1991.

III. PATTERNS OF POSTED RATES IN OUR SAMPLE OF LENDERS
We hypothesize that lenders post interest rates to send two types of signals to borrowers.
First, lenders may differ in the amount of risk they are willing to assume and may use posted rates
to signal the market of their willingness to accept risk. Second, from one month to the next,
individual lenders may find that they are in a position to make more or fewer loans, in which case
they may use posted rates to signal their willingness to accept loan applicants. We assume that
lenders consistently posting high rates relative to the market are signaling that they are willing to

clevelandfed.org/research/workpaper/index.cfm

accept more risky loans than lenders that consistently post below-market rates. We consider the
market rate to be the median interest rate advertised each month in each metropolitan area by our
sample of lenders. Each lender's median deviation from the market rate is considered to be its
long-term position in the market, which we assume to be a signal of its underlying type. Most of
the lenders in our sample (7 1 out of 87) posted rates that are within 118 point of the median rate
in their market for the majority of the 24 months covered. However, eight of the lenders posted
rates that were more than 118 point above the market for the majority of the months, and another
eight posted rates that were more than 118 point below. All eight high-rate lenders are
depository institutions (either commercial banks or thrifts), whereas only one of the low-rate
lenders is a depository institution, three are subsidiaries of depository institutions, and four are
independent mortgage banks.

In addition to using posted rates to signal their long-term position in the market, lenders
may vary their posted rates relative to their long-term position to signal their intention to accept
loan applications. Some lenders tend to move above or below their long-term positions for
extended periods, while others shift positions for relatively short durations. We use the number of
runs to describe the length of time that a lender shifts its position relative to the market. A
complete run extends from the time a lender moves above or below its long-term position to the
time it shifts from being above (below) to being below (above) its long-term position. Forty
lenders in our sample had so few runs that we can reject the hypothesis that their short-term
interest-rate changes are random (see table 1). These lenders tend to shift rates relative to their

clevelandfed.org/research/workpaper/index.cfm

long-term position infrequently and thus have prolonged periods when they are above or below
their long-term position.7

In addition to the length of time above or below their median position, lenders differ in the
magnitude of their rate changes relative to their long-term position. Some lenders tend to post
large changes in rates, while others post relatively small changes when they do shift from their
long-term position. We use the absolute value of non-zero deviations from each lender's longterm position to capture this sort of variation across lenders in the magnitude of short-term rate
changes. In table 1, lenders with a mean absolute deviation greater than 118 point are classified
as high variance lenders, while those with a mean absolute deviation below that are classified as
low variance lenders.
Table 2 reports the cross-classification of lendefs by both measures of short-term shifts in
posted rates. Forty percent of the lenders in our sample change relative position often, and the
magnitude of these changes is relatively small; 13.7 percent make frequent, large shifts in posted
rates; 28.7 percent make infrequent, small changes in their relative position; and the remaining
17.2 percent make infrequent, large changes in their relative position.

IV. POSTED RATES AS SIGNALS OF LENDERS' WILLINGNESS TO ACCEPT LOAN
APPLICATIONS
In a typical month, the majority of lenders advertise rates that are within 118 point of the
median rate posted in their market (see column 1 of table 3). However, almost 45 percent of

'The probability that the number of runs is random is calculated following Gibbons (1971). We classify
the lender as systematic in its shifks of posted rates if we can reject the null hypothesis of randomness at
the 5 percent confidence level.

clevelandfed.org/research/workpaper/index.cfm

lenders advertise rates that are at least 118 point above or below the market median, and in a
typical month, 16 percent advertise rates that are at least 218 point above or below the median.
This diversity of advertised rates suggests that lenders may be using their postings to signal the
market. One hypothesis is that when lenders want to increase the size of their mortgage
portfolio, they advertise rates that are lower than their normal market position. Borrowers
respond to the signal, and total applications to the lenders rise. The change in the quantity, and
possibly the quality, of the loan applications received, coupled with lenders' desire to increase the
size of their loan portfolios, may also mean a change in underwriting standards. As a result,
lenders' overall denial rates may also change.
In this section, we consider the relationship between short-term changes in posted rates
and four aspects of mortgage lending: total number of applications a lender receives, quality of
these applications, standards used in evaluating the applications, and overall denial rates. To
examine these relationships, we estimate the following:
(2)

Yit = ai Li + g~ M m + b~Rit+ eit.

Yit is a measure of mortgage lending activity for lender i in month t. The four measures of
mortgage lending activity considered are 1) total applications received, 2) the predicted denial
rate, which is used as a measure of the quality of loan applications received, 3) the difference
between actual and predicted denial rates, which is used as a measure of the standards being used
to evaluate loan applications, and 4) the actual denial rate. Rit is a vector of dummy variables
indicating the difference between lender i's posted interest rate in month i and the median interest
rate advertised in month i by lenders in the metropolitan area. Nine categories are included that

clevelandfed.org/research/workpaper/index.cfm

indicate how far each lender is from the median rate; the median market rate is the omitted
category. Li is a vector of dummy variables for each lender, included to control for normal
mortgage lending activity and deviation from the market rate, and M, is a vector of dummy
variables for each metropolitan area M in month t, included to control for marketwide changes in
mortgage lending activity and interest rates. The applications regression is estimated using
ordinary least squares; all other regressions are estimated using weighted least squares, where the
weights are the number of applications received.
Parameter estimates for the interest-rate variables in equation (2) are presented in table 3.
Since these regressions control for lender-specific and market/month-specific effects, these
coefficients can be interpreted as the change in a lender's monthly application flows, denial rates,
quality of applications, or lending standards that is associated with a change in the lender's relative
position in the market, as measured by its deviation from the median market rate.
Our estimates are largely consistent with the argument that lenders use posted rates to
signal their willingness to accept loan applications and that the market responds to these signals.
Application flows increase significantly when lenders lower their advertised rates relative to their
normal position in the market. The point estimates indicate that the elasticity of applications with
respect to the posted rate is quite high: Applications increase by about 20 percent for every 118
point reduction in posted rates relative to the lender's normal market position. Applications also
decrease when lenders raise their posted rates relative to their normal position. However, the
change in applications is relatively small and is significant only when the rate is 118 to 218 point
above the normal position.

clevelandfed.org/research/workpaper/index.cfm

The quality of loan applications, measured by predicted denial rates, also responds to
changes in posted rates. Our estimates indicate that quality increases when lenders lower their
posted rates relative to the market and to their normal position in the market. The quality of loan
applications also falls off when posted rates rise, suggesting that when a lender increases its
posted rates, only those applicants with the least probability of acceptance continue to apply.
It also appears that the underwriting standards applied to loan applications change when
lenders alter their posted rates, particularly when the new rates are 218 to 318 point above or
below the market, given their normal position relative to the market. When lenders lower their
rates, the gap between their actual denial rate and their predicted denial rate shrinks, indicating
that for a given quality of loan application, lenders are denying fewer loans and hence are
lowering their acceptance standards. This behavior is consistent with lenders dropping their rates
to augment the size of their loan portfolios. Similarly, when lenders raise their rates 218 to 318
point above the market, given their normal position, their denial rates increase relative to the
quality of applications received, indicating that their acceptance standards are higher.
Finally, we find that actual denial rates rise when lenders increase their posted interest
rates and fall when lenders decrease their rates. Again, this effect is significant only if rates are
218 to 318 point above or below the market.

In summary, when lenders lower (raise) their rates relative to their normal position in the
market, they also tend to lower (raise) their acceptance standards. These two effects work in the
same direction in regard to application flow: The number of applications received increases
(decreases) when lenders lower (raise) their posted rates relative to their normal position in the

clevelandfed.org/research/workpaper/index.cfm

market. These findings are consistent with lenders using short-term changes in posted rates to
signal the market of their intention to accept more or fewer loan applications, and the market
responds to these signals. Interestingly, the average quality of applications received increases
when lenders post lower rates and decreases when lenders post higher rates, though it is not
immediately obvious why this should be true. One would expect that lowering interest rates and
acceptance standards would increase applications from both high- and low-risk applicants, and
that increasing rates should reduce applications from both groups. From our results, it appears
that the relatively low-risk applicants are more sensitive to changes in posted rates.

V. POSTED RATES AS SIGNALS OF LENDERS' WILLINGNESS TO ACCEPT RISK

In addition to using posted rates to signal their desire to increase or decrease the size of
their loan portfolios, lenders may use this channel to signal their willingness to accept risk. If the
market is composed of high-risk and low-risk lenders, lenders may use posted interest rates to
signal to borrowers which type they represent. As discussed in section TI, we consider high-rate
lenders to be those with posted rates that are more than 118 point above the market for the
majority of the 24 months in our sample, and low-rate lenders to be those with posted rates that
are more than 118 point below. In our sample, we identify eight high-rate and eight low-rate
lenders.
Table 4 presents mean characteristics for high-, middle-, and low-rate lenders in our
sample. The first panel presents the means weighted by the number of applications received by
each lender. The entries in the first row indicate that 3.48 percent of all applications received by

clevelandfed.org/research/workpaper/index.cfm

low-rate lenders were rejected, compared with 9.98 percent of those received by high-rate
lenders. In the bottom panel, the lender, rather than the application, is the unit of analysis used in
calculating the means. The entries in the first row indicate that the average denial rate for lowrate lenders is 5.98 percent, compared with 13.24 percent for high-rate lenders. The difference
between the means for lowlmiddle- and middlehigh-rate lenders, and F-tests of the significance of
these differences, are presented in table 5.
Our estimates are largely consistent with the argument that lenders use posted rates to
signal the market of their willingness to accept risk, and that the market is responding to these
signals: The quality of applications received by the low-ratellow-risk lenders is significantly
higher than for the middle- or high-rate lenders. Beyond this, low-rate lenders tend to sell off a
larger portion of the loans they originate, to apply less stringent underwriting standards, and to
deny fewer loan applications than middle- or high-rate lenders. In general, differences between
low-rate and middle-rate lenders appear to be considerably larger and more significant than
differences between middle- and high-rate lenders.
Since our classification of lenders as high or low rate is somewhat arbitrary, we also
examine the correlation between lenders' median deviation from the market rate and various
measures of mortgage lending activity. These correlations are presented in table 6. The simple
correlations, presented in the first column, are consistent with the differences in means presented

in tables 4 and 5. The quality of loan applications is lower for high-rate lenders. Ln addition,
these lenders tend to have higher denial rates, high loan acceptance standards, and a greater
number of loans held in their portfolios than do low-rate lenders.

clevelandfed.org/research/workpaper/index.cfm

As reported in table 1, all of the high-rate lenders in our sample are depository institutions
(either commercial banks or thrifts), while the low-rate lenders are either mortgage subsidiaries of
banks and thrifts or independent mortgage banks. It may be that these different types of
institutions generally service different types of customers and apply different standards in
evaluating loan applications, independent of any variations across them in their advertised interest
rates.'

The second column of table 6 presents partial correlations between lenders' median

deviation from the market rate and our measures of lending activity, controlling for the type of
institution. These partial correlations indicate that the differences related to posted rates that
were observed across all lenders are not due solely to differences in institutional type. Even after
controlling for type of institution, high-rate lenders have higher denial rates, lower average quality
of applications, and higher loan acceptance standards. The only difference that seems to be
related solely to differences in institutional type is the percentage of loans sold.

VI. CONCLUSIONS
The mortgage lending market has come under increasing scrutiny in recent years amid
allegations that lenders are underserving some neighborhoods and allowing race to enter the
lending decision. We investigate neither issue in this paper. However, in previous work (Avery,
Beeson, and Sniderman [1993a, 1993b, and 1994b1) on these topics, we argue that the behavior
of both lenders and borrowers needs to be analyzed more carefully to truly understand the

his finding is consistent with results reported by Benjamin, Heuson, and Sirmans (1994) from a
sample of South Florida mortgage lenders.
14

clevelandfed.org/research/workpaper/index.cfm

mortgage credit process. Considering the role of interest rates in the market is a step in that
direction.

In this paper, we discuss how lenders might use posted lending terms to signal both their
eagerness to accept new loan applications and their lending standards relative to other lenders in
their market. We demonstrate that lenders who lower their posted rates relative to their own
normal market position indeed attract more applicants. At the same time, lenders who lower their
rates also appear to loosen their credit standards, which should reinforce that pattern.
We also find that better quality applicants are more likely to apply to low-rate lenders and
that these lenders tend to sell off a larger portion of the loans they originate, to apply less
stringent underwriting standards, and to deny fewer loan applications than do middle- or high-rate
lenders. In our sample, the low-ratebow-risk lenders are generally independent mortgage banks
and the mortgage subsidiaries of commercial banks and thrifts. The high-ratelhigh-risk lenders
tend to be commercial banks and thrifts. These lenders may be playing different roles in their
respective markets.
To our knowledge, this paper is the only empirical examination of interest rates as signals
in the mortgage lending market. If a dataset such as HMDA could be assembled that also
included the actual credit terms of loan applications, many interesting questions could be
explored. One issue is the extent to which posted rates accurately signal a lender's transaction
prices. Another line of research would be the development of matching models to gauge more
precisely how the price of credit is related to credit risk in this market.

clevelandfed.org/research/workpaper/index.cfm

REFERENCES

Avery, Robert B., Patricia E. Beeson, and Mark S. Sniderman. 1993a. "Accounting for
Racial Differences in Housing Credit Markets." Proceedings of a Conference on
Discrimination and Mortgage Lending, U.S. Department of Housing and Urban
Development, forthcoming. Also, Working Paper 9309, Federal Reserve Bank of
Cleveland (August).

. 1993b. "Lender Consistency in Housing Credit Markets." Proceedings of the
1993 Conference on Bank Structure, Federal Reserve Bank of Chicago, pp. 339-58.
Also, Working Paper 93 10, Federal Reserve Bank of Cleveland (August).

. 1994a. "Underserved Mortgage Markets: Evidence from HMDA." Working
Paper 9416, Federal Reserve Bank of Cleveland (November).
. 1994b. "Cross-Lender Variation in Home Mortgage Lending."

Economic
Review, Federal Reserve Bank of Cleveland, vol. 30, no. 4 (Quarter 4 1994), pp. 15-29.

Benjamin, John D., Andrea J. Heuson, and C.F. Sirmans. 1994. "Are Depository
Institutions and Mortgage Bankers Different? Evidence from the South Florida Market."
Journal of Housing Research, vol. 5, no. 1, pp. 139-70.
Canner, Glenn B., and Delores S. Smith. 1991. "Home Mortgage Disclosure Act:
Expanded Data on Residential Lending." Federal Reserve Bulletin, vol. 77
(November), pp. 859-8 1.

.

1992. "Expanded HMDA Data on Residential Lending: One Year Later."
Federal Reserve Bulletin, vol. 78 (November), pp. 801-24.

Gibbons, Jean D. 1971. Nonparametric Statistical Inference. New Y ork: McGraw-Hill.

clevelandfed.org/research/workpaper/index.cfm

Table 1: Distribution of Lenders

Commercial Banks
and Thrifts

Mortgage Subs of
Banks and Thrifts

Independent
Mortgage Banks

Total

19

17

87

Median Deviation from Market
Rate
High (+ 118 point or more)

8

Mid (-118 to +1/8 point)

42

Low (-118 point or more)

1

Shifts in Relative Market
Position
Systematic (prolonged) shifts

23

Random (frequent) shifts

28

Variance of Deviation from
Market Rate

Low (average absolute
deviation < 118 point)

33

High (average absolute
deviation > 118 point)

18

Total
Source: Authors' calculations.

51

clevelandfed.org/research/workpaper/index.cfm

Table 2: Short-term Lender Types

Changes in Relative Market Position

Low Variance
(average absolute deviation
< 118 point)
Percent of lenders
High Variance
(average absolute deviation
> 118 point)
Percent of lenders

Source: Authors' calculations.

Systematic (Prolonged)
Shifts

Random (Frequent) Shifts

25

35

28.7%

40.2%

15

12

17.2%

13.7%

clevelandfed.org/research/workpaper/index.cfm

Table 3: Relationship between Month-to-Month Changes in Posted Interest Rates and Application
Flows, Quality of Loan Applications, Lender Standards, and Denial Rates

Parameter Estimates
Deviation from Market Median

More than 318 point above

1

'

Percent of
Lenders

Total
Applications

Predicted
Denial Rate
(Quality)

Actual
Denial Rate

3.96

-3.076
(4.17 1)

-0.006
(0.264)

0.330
(1.175)

( 1.093)

ActualPredicted
Denial Rate
(Standards)
0.336

218 to 318 point above

8.02

-4.208
(3.158)

0.526"
(0.165)

2.840"
(0.737)

2.3 13"
(0.685)

118 to 218 point above

15.69

-4.608"
(2.580)

0.150
(0.126)

0.257
(0.561)

0.107
(0.522)

0 to 118 point above

18.37

-3.289
(2.540)

0.193
(0.123)

0.07 1
(0.546)

-0.122
(0.508)

0

14.65

0 to 118 point below

23.58

1.338
(2.437)

0.004
(0.264)

0.1 13
(0.529)

0.109
(0.492)

118 to 218 point below

11.70

7.746"
(2.848)

-0.1 17
(0.142)

-0.53 1
(0.633)

-0.414
(0.5 89)

218 to 318 point below

3.05

14.549"
(4.500)

-0.436"
(0.263)

-2.233b
(1.170)

- 1.797"
(1.088)

More than 318 point below

0.97

33.936"
(7.177)

-0.770~
(0.359)

- 1SO7
(1.598)

-0.737
(1.486)

R~

0.8668

0.6560

0.5015

0.4635

Mean of dependent variable

39.26

11.05

12.73

1.68

Number of observations

2880

2427

2427

2427

Regressions include dummy variables for each lender and for each month*MSA combination.
Note: (standard errors) " " indicate significant at 1 percent, 5 percent, and 10 percent levels of confidence,
respectively.
Source: Authors' calculations.

clevelandfed.org/research/workpaper/index.cfm

Table 4: Mean Application Characteristics by Long-term Market Position (High-, Middle-,
and Low-Rate Lenders)

-

Weighted by Number of Applicants:
Low-Rate

Middle-Rate

High-Rate

Low-Rate

Middle-Rate

High-Rate

Actual denial rate
Predicted denial rate
Actual-predicted denial rate
Percent of loans sold
Percent minority applicants

Unweighted:

Actual denial rate
Predicted denial rate
Actual-predicted denial rate
Percent of loans sold
Percent minority applicants

Average number of applicants
Number of lenders
Source: Authors' calculations.

clevelandfed.org/research/workpaper/index.cfm

Table 5: Differences in Means across Lender Types

Weighted
MiddleLow

HighMiddle

Unweighted
MiddleLow

HighMiddle

Actual denial rate
Predicted denial rate
Actual-predicted denial rate

3.45
(1.1227)

Percent of loans sold
Percent minority applicants

0.2 1
(0.0087)

Average number of applicants

* Indicates means are significantly different at the 10 percent level of confidence. F-statistics are in
parentheses.
Source: Authors' calculations.

clevelandfed.org/research/workpaper/index.cfm

Table 6: Correlations between Application Characteristics and Lenders' Median Deviation
from the Market Interest Rate

Correlation Coefficients
Simple Correlation
Coefficient

Partial - Controlling
for Lender ~ y p e '

Actual denial rate

0.2764*

0.3143"

Predicted denial rate

0.2493"

0.221 l *

Actual-predicted denial rate

0.243 1*

0.2956*

Percent of loans sold

-0.4748*

-0.1013

Percent minority applicants

-0.1529

-0.0464

Average number of applicants

0.0 1.54

-0.0834

* Indicates significantly different from zero at the 10 percent level of confidence.

' Commercial bank or thrift, subsidiary of a commercial bank or thrift, and independent mortgage bank.
Source: Authors' calculations.