-
Originator Performance, CMBS Structures and Yield
Spreads of Commercial Mortgages∗
Sheridan Titman
McCombs School of Business
Department of Finance
University of Texas at Austin
Austin, TX 78712-1179.
Sergey Tsyplakov
Moore School of Business
Department of Finance
University of South Carolina
Columbia, SC 29208
Current Draft: May 28, 2008
∗Authors’ e-mail addresses are [email protected] and
[email protected] respectively. We
would like to thank seminar participants at the University of
Texas at Austin, UCLA, the University of South
Carolina, and the 2007 Summer Real Estate Symposium, and,
especially, Jefferson Duarte, Tim Koch, Steve
Mann, Eric Powers and Donghang Zhang. We also thank Renjie Wen
for excellent research assistance.
-
Originator Performance, CMBS Structures and Yield Spreads of
Commercial
Mortgages
Abstract
This paper examines information and incentive problems that can
exist in the market
for conduit mortgages, which are commercial mortgages placed in
pools that are repackaged
and sold as CMBS. We find that conduit mortgages that are
originated by institutions
with negative stock price performance in the quarters just prior
to the origination date
tend to have higher credit spreads and default more than other
mortgages with similar
characteristics. This evidence is consistent with reputation
models that suggest that poorly
performing originators have less incentive to expend resources
evaluating the credit quality
of prospective borrowers. We also find that the
originator/performance effect is stronger
when the originator of the mortgage is also the lead underwriter
of the CMBS and that the
time between the origination date and the CMBS offering is
shorter for originators that are
stock price losers.
-
1 Introduction
Securitization involves pooling cash flow claims from financial
instruments to form publicly
traded securities. Financial instruments that are commonly
securitized include residential
and commercial mortgages and commercial and consumer loans,
which historically, were
both originated and held by banks, savings and loans and
insurance companies.
With the introduction of securitization, the evaluation and risk
bearing functions of
financial institutions can be split. The institution that
initially underwrites the loan, and
evaluates the borrower, need not be the same institution that
ultimately bears the risks
associated with holding the loan. The advantage of separating
these functions is that the
loan originations are best performed by institutions with good
local knowledge, while risk is
most efficiently borne by large internationally diversified
portfolios. Offsetting this advantage
are potential incentive/information problems that exist when an
originator with better local
knowledge sells mortgages to less informed investors. In
particular, there may be a tendency,
on the part of the originator, to expend too little effort in
the evaluation of individual loans.
Due to the repeated nature of the securitization business, such
a tendency is not expected to
be severe during normal circumstances because originators of
individual loans are expected
to be concerned about their reputation for originating high
quality loans. However, there
may be a tendency to reduce origination expenses when the
originator is doing poorly and
is more concerned about its short-run profitability than its
reputation.
This paper examines these incentive/reputation issues within the
context of what are
called conduit mortgages, which are commercial mortgages that
were originated with the in-
tention of placing them in Commercial Mortgage Backed Securities
(CMBS). The commercial
mortgage backed securities market was introduced in the early
1990s and grew very rapidly,
perhaps because of the advantages associated with liquidity and
better risk sharing;1 by
2006 there were more than $600 billion in CMBS bonds
outstanding.2 However, the CMBS
1See Riddiough and Polleys (1999), Riddiough (2002), and Ambrose
and Sanders (2003).2In 2005 more than $165 billion in CMBS was
issued in the United States, which represented more than
20% of the overall commercial mortgage market. These numbers are
quoted from a January 8, 2006 press
release of the Commercial Mortgage Securities Association
(CMSA).
1
-
market slowed considerably in the latter half of 2007, in part,
reflecting concerns about the
origination/underwriting process induced by residential
sub-prime mortgage problems.3
To understand the information issues that arise in the
origination of commercial mort-
gages it is useful to contrast the commercial mortgage market
with the market for residential
mortgages. In contrast to commercial mortgages, which are not
always securitized, most resi-
dential mortgages are now packaged into mortgage backed
securities. One difference between
these markets is that for residential mortgages originators tend
to use what are known as
credit scoring systems, which are purely mechanical systems that
map quantitative informa-
tion about the borrower and the property into a credit score
that determines whether the
loan is offered. By using a credit scoring system banks ignore
soft information (i.e., infor-
mation that cannot easily be quantified), which plays an
important role in the evaluation of
commercial mortgages. In addition to evaluating hard facts like
loan to value and coverage
ratios, the originator of commercial mortgages evaluates the
quality of the property, its al-
ternative uses,4 and the incentives and reliability of the
owners. This type of information
is likely to be costly to quantify, which means that the
ultimate mortgage investors, who
see only the hard information, must to a large extent rely on
the originator’s judgment and
reputation. A reputable originator, who is expected to carefully
investigate the mortgages
and exercise good judgment, may be able to sell mortgages for a
premium relative to less
reputable originators who are likely to expend less resources
evaluating the mortgages.
There is a large theoretical literature that describes the
tradeoffs between the short-
term benefits of exploiting one’s reputation, e.g., selling
mortgages without bearing the
investigation costs, and the long-term costs of operating in the
future without the benefit of
a favorable reputation.5 These models suggest that in situations
where the originator has a
long horizon, i.e., uses a very low discount rate to evaluate
this tradeoff, there is a tendency to
make choices that are likely to help its reputation. However, a
more short-sighted originator
may make choices, like reducing investigation costs, that help
its current earnings at the
3See, for example, Prudential Real Estate Investors Quarterly
Report for October 2007.4Benmelech, Garmaise, and Moskowitz (2005)
presents evidence that the redeployability of commercial
real estate affects mortgage spreads.5Klein and Leffler (1981),
Shapiro (1983), and Allen (1984) provide models where buyers cannot
observe
product quality prior to purchase. In these multi-period models,
sellers forgo the short-term cost savings
associated with reducing quality in order to maintain their
reputations and ability to credibly sell high
quality goods in the future.
2
-
expense of its reputation and future earnings.6
Our empirical analysis of these issues is based on the idea that
an originator’s past stock
price performance is an indicator of its incentive to make
short-sighted choices that help its
current earnings at the expense of its reputation. As a result,
relative to the hard information
that is available to CMBS investors, the mortgages originated by
poorly performing institu-
tions may be somewhat riskier. There are a number of reasons why
this may be the case.
First, poorly performing originators may be more likely to
approve mortgages without the
appropriate due diligence, thereby letting some relatively risky
mortgages pass through their
screens. Second, on the margin, a poorly performing originator
may be less willing to forego
origination fees by rejecting mortgage applications when its
investigation reveals potentially
unfavorable information about the borrower or the property.
Third, it is plausible that
poorly performing originators are under pressure to sell riskier
mortgages and place them
in CMBS deals even though these mortgages were not intended
initially to be securitized.
Finally, a poorly performing originator may attract less credit
worthy mortgagors, who are
aware of their weaker screening standards. As a result, if the
buyers of conduit mortgages are
aware of these tendencies they will require higher credit
spreads, conditional on observable
characteristics, on the mortgages they acquire from poorly
performing originators.7
Consistent with this argument we find that originators that
experienced large stock price
declines in prior quarters tend to originate mortgages with
wider spreads, even after con-
trolling for a variety of property and mortgage characteristics,
as well as originator fixed
effects and monthly time effects, which control for the fact
that some institutions specialize
in riskier mortgages and that economy-wide shocks change the
riskiness of real estate from
month to month. However, while this evidence supports the
reputation hypothesis, there
are other potential explanations. In particular, it is possible
that the direction of causality
is the reverse of what we suggest. Specifically, it could be the
case that some originating
institutions experience lower stock returns because of their
focus on geographical regions or
6Maksimovic and Titman (1991) develop a model, based on Myers
(1977), where the firm’s cost of raising
capital is higher when it is financially distressed. Because of
its higher discount rate, the firm has less
incentive to improve product quality that cannot be initially
observed. One can generate similar results
within the context of a model where the manager has shorter
horizon following poor performance because
of career concerns.7See, for example, Titman and Trueman (1986)
and Khanna, Noe, and Sonti (2006) which provide models
where better quality underwriters attract better quality
issuers.
3
-
property types which have become particularly risky (relative to
the entire market), and thus
originate mortgages with higher relative spreads that reflect
the higher relative risk. It is
also possible that there is a reputation channel that is very
different than what we propose.
For example, Hubbard, Kuttner and Palia (2002) suggest that
financial institutions that are
doing poorly tend to be less concerned about their long-term
reputation and exploit their
borrowers by charging higher rates on their commercial
loans.
To distinguish between these alternative hypotheses we run
additional cross-sectional
test. We find that the difference in spreads between stock price
losers and non-losers is
wider during periods when whole financial industry experiences
market downturns. This
evidence is consistent with the idea that scrutiny of
origination quality can weaken during
market downturns which allows potentially distressed
institutions to shift their origination
practices, and less consistent with hypothesis that stock prices
decline in anticipation of
risk increase of certain regions of real estate markets. Our
most direct evidence that the
mortgages of stock price losers are riskier comes from an
analysis of the default histories of
the mortgages. We find that the mortgages originated by stock
price losers default more
often, suggesting that the observed higher spreads reflect
higher default risk rather than the
exploitation of borrowers. In addition, we find that the time
lag between when a mortgage
is originated and the selling date of the CMBS deal that
includes the mortgage is shorter
for mortgages originated by originators who had poorer stock
price performance. This last
finding suggests that poorly performing originators may be
anxious to sell the mortgages
they originate before negative information about the mortgages
is revealed. Finally, we
provide evidence that suggests that the ratings agencies act as
though the past stock returns
of the originators provide information about a mortgage’s risk.
Specifically, we find that the
ratings agencies require higher subordination levels (a lower
percent of the issued bonds are
rated AAA) for CMBS issues that include more mortgages
originated by underperforming
originators.
We also examine the loan to value ratios of the mortgages, which
may be influenced by
the originator’s past stock returns under both of the
alternative explanations. For example,
if we believe stock price losers have riskier clients, or that
they exploit them more, then we
might expect to observe lower loan to value ratios on the
mortgages they originate. However,
we find no significant relation between loan to value ratios and
the originator’s past stock
returns.
4
-
Finally, we uncover evidence that provides support for the idea
that the degree of outside
scrutiny plays an important role in the determination of credit
spreads. Specifically, in our
data set, about a third of the mortgages are originated by the
same institution that later
becomes the lead underwriter of the CMBS in which the mortgage
is sold. When this is the
case, the mortgage may be scrutinized less before it becomes a
part of a CMBS and ultimately
sold to investors. We find that in these cases the past stock
returns of the originator have a
stronger influence on mortgage spreads.
Although our focus is on the CMBS market, our analysis relates
to the broader litera-
ture that considers the relation between the reputations of
financial intermediaries and the
pricing of equity and debt issues. This literature includes
theoretical papers that examine
how the reputation of underwriters relate to the quality of the
corporate securities they
issue,8 as well as an empirical literature that explores the
cross-sectional relation between
various measures of underwriter reputation and security
pricing.9 However, in contrast to
this literature, which views the reputation or prestige of the
underwriting institution as a
permanent attribute,10 our focus is on how the credibility of an
institution can change over
time. Indeed, our empirical tests include originator fixed
effects and thus control for differ-
ences in their “average” reputations, and focus on how the
credibility of originators change
when their stock return is lower.
The organization of this paper is as follows: next analyzes the
CMBS securitization
process. Section 3 describes the data. Our main results,
including robustness checks, are
discussed in Section 4. Additional tests are presented in
Section 5. Section 6 concludes the
paper.
8For example, in Pennacchi (1988), a bank’s ability to sell
loans to investors depends on investors’
perceptions of the bank’s ability to monitor loans they sell. In
addition, in Chemmanur and Fulghieri
(1994), the reputation of a financial intermediary mitigates the
moral hazard problem associated with the
incentive to lower the underwriting standards.9Beatty and Ritter
(1986) and Carter and Manaster (1990) find that the prestige of
investment banks
is associated with the pricing of IPOs. Michaely and Shaw (1994)
document that IPOs underwritten by
reputable investment banks experience less underpricing and also
show that they perform better in the long
run. Livingston and Miller (2000) and Fang (2005) finds that
debt issued by more prestigious underwriters
have lower yields.10Carter and Manaster (1990) measure prestige
of investment banks its place in the hierarchy of bank names
in "tombstone announcements." Michaely and Shaw (1994) and Fang
(2005) use a measure of investment
bank reputation based on bank’s relative size and/or market
share.
5
-
2 The CMBS Securitization Process
In this section we describe institutional details of the process
through which a typical CMBS
security is created. The process starts with a financial
institution (the originator) that pro-
vides a mortgage that is collateralized by commercial property.
If the mortgage is originated
with the intention of including it in a CMBS, it is called a
conduit loan and the originator
has to follow certain loan origination guidelines in assessing
the mortgage risk. The origi-
nator has to assess a variety of risk factors related to
property type, location, the physical
quality of the property, lease structure, the quality of the
management team and the quality
of the tenants.11 Some of these risk factors, such as property
type, size of the property and
the current lease structure are either observable or relatively
easy to assess. However, in
addition to this hard information, they must make more ambiguous
assessments of softer
information, such as local competition and the quality of the
property and its tenants. This
soft information can be expensive to obtain and may be difficult
to verify.
When the mortgage terms are finalized the originator transfers
(or sells) the mortgage
along with other mortgages to a CMBS underwriter. Normally there
is a time lag between
when the mortgage is originated and when it is transferred to a
CMBS underwriter. The
underwriter screens the mortgages and creates a pool of
mortgages from different origina-
tors. Using this pool as collateral the underwriter issues bonds
with different seniorities
(or tranches) that are sold to investors.12 Often the
underwriter includes the mortgages it
originated in the pool, in which case, there is no third party
that screens the mortgages.
After the pool composition is finalized, but before the
securities are issued, the credit rat-
ing agencies analyze the portfolio of mortgages in the pool,
considering the above mentioned
characteristics of the mortgage as well as portfolio
characteristics, like the property type
composition and geographical diversification. In many cases the
credit agencies reexamine
the quality of the mortgage origination process and conduct
individual site inspections.13
Based on their assessment of the risk factors, rating agencies
determine the subordination
levels for each credit rating. For example, AAA subordination is
the percentage value of the
total CMBS pool that are in securities rated below AAA. Since a
higher subordination level
reduces the default risk of the bonds, subordination levels for
each rating category will be
11See Fabozzi et al (2000).12See Riddiough (2002) for a
description of the underwriters role in the CMBS markets.13See
Fabozzi and Jacob (1998).
6
-
higher when the pool of mortgages is deemed to be riskier.
3 Data Overview
Our data set, which was provided by Standard & Poors,
includes information on over 19,000
commercial mortgages that were originated between 1992 and 2002.
As reported in Panel 1
of Table 1, very few mortgages were originated prior to 1995, so
we limit our study to the
1995 to 2002 period. Most of the mortgages in our sample are
conduit mortgages, issued
specifically for inclusion in a commercial mortgage backed
security (CMBS).14 The data set
includes detailed information on characteristics of the
mortgaged properties along with the
mortgage contract specifications and information on mortgage
originators and CMBS deals.
Our data set also includes the dates when the mortgage is
originated as well as the dates
when the CMBS that includes the mortgage is issued. The time
between these dates, i.e.,
the "time lag," average 159 days with a standard deviation of
141 days. The time lag for
some mortgages is several years.
For most mortgages we have information on the CMBS deal to which
the mortgage was
subsequently sold. Specifically, we have data on 94 CMBS deals
that include a total of
14,420 mortgages.15 ,16 For these deals we have information on
the lead underwriters, which
structure the securities and sell them to the CMBS investors.
The lead underwriters are
often the originator of the largest number of mortgages in a
given deal, but may also be a
third party, i.e., not an originator of any mortgage in the
deal. As we show in the summary
statistics, originators that also serve as underwriters of CMBS
deals tend to include more
than 50% of their own mortgages in those deals.
3.1 Property Characteristics
The data set also includes information on the mortgaged
properties. The information we
use includes the property type and the appraised property value
as well as its annual net
14Our sample also includes a small number of mortgages that were
included in CMBS pools more than 2
years after origination.15For more than 4,000 mortgages we
cannot identify the CMBS deals to which the mortgages were
sold.16Typically, the CMBS deals include between 100 and 200
mortgages. The largest number of mortgages
in a deal is 591 and one deal has only one mortgage.
7
-
operating income at the time of the origination.17 Summary
statistics describing the proper-
ties are presented in Panel 3 of Table 1. The property types
include multi-family apartment
complexes, unanchored retail, anchored retail, medical offices,
industrial, warehouse, mo-
bile home parks, office buildings, properties of mixed use,
limited service hotels, full service
hotels, and self storage. The data set also includes information
about occupancy of the
property and the year it was built or renovated. These two
observations are missing for
about 15% of the mortgages in the data set.
3.2 Mortgage Characteristics
The data includes the following financial information for
individual mortgages: origination
date; mortgage rate; loan to value ratio; whether the mortgage
is non-amortizing, amortizing,
or semi-amortizing; and the maturity of the mortgage. The loan
to value ratio (LTV), which
is generally between 60% and 80%, is measured as the loan amount
divided by the appraised
value of the property. Non-amortizing mortgages represent
approximately two-thirds of the
mortgages with the rest being amortizing and semi-amortizing
mortgages, where the loan
amortization rate is defined as 1−Principal Value at
MaturityInitial Principal Value (See Panel 3 of Table 1). The
majorityof the mortgages have 10 year maturities and, due to
prepayment penalties, are effectively
not prepayable.
3.3 Originator Characteristics
The mortgages in our final sample were originated by 50
different institutions. Mortgage
originators include large commercial banks, investment banks,
insurance companies, and
financing arms of large companies (e.g. GMAC). In Panel 4 of
Table 1, we list the 30
largest originators that have at least 100 mortgages in the data
set. Mortgages originated
by these institutions constitute about 97% of the mortgages in
the data set. We also include
information on average spread, LTV ratios and delinquency rates
across the originators. As
reported in the summary statistics, seven originators issued
more than 1,000 mortgages each.
Mortgage LTV ratios do not vary significantly across
originators, but the delinquency rates
17The Net Operating Income (NOI) is defined as gross annual
revenue less maintenance and other oper-
ational expenses before taxes and depreciation for the 12 month
period prior to the mortgage origination
date.
8
-
do vary across originators.
The originator characteristic that we are most interested in is
their cumulative stock
returns in periods prior to the mortgages they originate. Panel
5 of Table 1, reports cu-
mulative stock returns (and standard deviations) across
originators measured during two
look-back windows of three and six month prior to mortgage
origination. Given that our
sample periods includes the Asian currency crises, the Russian
default, as well as the period
after 9/11, there is a time-series variation as well as
cross-sectional variation in these stock
returns.
4 Main Regression Analysis
4.1 Cross-Sectional Regression Specification
As we mentioned in the introduction, our goal is to estimate the
extent to which the orig-
inators’ past performance influences the characteristics of the
mortgages they originate. In
particular, we will examine the relation between originators’
past stock returns and the
spreads on the mortgages they originate. In these tests we
define mortgage spreads as the
difference between the mortgage rate and the rate on Treasury
bonds with the same maturity
as the mortgage, observed on the mortgage origination date.
The first regression we estimate, described below, regresses
mortgage spreads on several
indicator (Dummy) variables describing the originators’ past
stock returns, along with control
variables that correspond to property and mortgage
characteristics.
Spread = intercept+ α(dummy=1 if originator is "stock price
loser")
+X
βi(property characteristics variables)
+X
γi(mortgage characteristics variables)
+ (property type dummy variables)
+ (originator dummy variable)
+ (origination time dummy variables) + .
(1)
We call the indicator variables for poorly performing
originators "stock price loser" dum-
mies and estimate a series of regressions with two different
stock price loser dummies, defined
as follows:
9
-
1) Dummy variable=1, if the originator’s stock return is less
than -10% during the 3
months prior to the mortgage origination date [-3mo, 0] and zero
otherwise.
2) Dummy variable=1, if the originator’s stock return is less
than -10% during the 6
months prior to the mortgage origination date [-6mo, 0] and zero
otherwise.
We choose negative 10% return as a cutoff because this return is
approximaty one stan-
dard deviation from avergae. See Panel 5 of Table 1. Panel 6,
which reports the stock return
characteristics of originators, reveals that between 10% and 11%
of the mortgages in our
sample were originated by institutions that realized stock price
declines of at least 15% in
the prior periods. There is significant time-series variation
across quarters in the fraction of
originators that are "stock price losers." For example, more
than 40% of the mortgages were
originated by stock price losers (prior 3mo window) in both the
fourth quarter of 1998 (the
period after the Russian default) and the fourth quarter of 2001
(the period after 9/11). In
contrast, in the first and second quarters of 1996, no mortgages
were originated by stock
price losers.18
We include control variables that describe characteristics that
are likely to proxy for the
riskiness of the properties. These variables include the
NOI/Value ratio, which proxies for
the expected growth of net operating income,19 and the value of
the property to capture size
effects.20 We also include property type dummies to control for
differences in the risk for
each type of property. In addition we include variables that
describe the mortgage terms,
which include the LTV ratio, the loan amortization rate, and the
mortgage maturity. The
loan amortization rate and the mortgage maturity measure how
fast the loan is paid off.
To control for possible differences across originators we use
originator fixed effects, and to
18Alternative measures for the periods when originators perform
poorly may include periods with low ROE,
EPS ratios or capitalization ratios as well as periods of debt
ratings reduction. However, these alternative
indicators may be less suitable given the heterogeneity of
originators. The concern is that we cannot directly
compare financial ratios across originators that include not
only banks and financial institutions, but also
insurance companies and arms of large corporations (GMAC). Using
periods with large stock return losses
as an indicator for poor performance is not subject to such a
concern.19Properties with larger NOI/Value ratios are likely to
have higher payouts and lower NOI growth in the
future. As shown in Titman, Tompaidis and Tsyplakov (2004),
NOI/Value should be positively related to
credit spreads.20First, there may be economies of scale
associated with the transaction costs of providing a mortgage.
Second, more reliable individuals may be acquiring the larger
properties. Finally, the larger properties may
have less risky cash flows because they may be more diversified
and the owners may have more market power.
10
-
control for macro changes in interest rates and risk we include
time dummies that correspond
to the different origination months.
4.2 Cross-sectional Results
4.2.1 The Relation Between Originators’ Prior Stock Returns and
Mortgage
Spreads
The estimates of these regressions, which are reported in Table
2, are consistent with our
hypothesis that poorly performing institutions originate riskier
mortgages. Panel 1 of this
table reports regressions that include monthly dummies and Panel
2 reports the regressions
that include macro-economic variables and annual dummies. In
both cases the coefficient of
the stock price dummy variables are positive and statistically
significant, which reveals that
after controlling for mortgage and property characteristics, as
well as originator and time
fixed effects, originators that experienced poorer stock price
returns in prior quarters tend
to issue mortgages with wider spreads. The coefficient of the
"stock price loser" dummy is
7 and 8 b.p., for look-back window of 3 and 6 months
respectively.21 If one believes that the
increase in average industry credit spreads that are associated
with larger fraction of stock
price losers is due to the originators providing less analysis
on the mortgages they initiate
21In an alternative specification, instead of using monthly time
dummies, we introduce variables that
control for changes in marco-economic credit environment and
changes in real estate fundamentals. The
variables we include are shown to account for changes in credit
spreads of commercial mortgages (e.g. Maris
and Segal. (2002) Titman,Tompaidis and Tsyplakov (2005))
including: spread of AAA-rated corporate
bonds; slope of the treasury curve, which is measured as Yield
of 10-year Treasury minus Yield of 1-year
maturity Treasury; cumulative return of NCREIF for prior four
quarters, average rate of commercial real
estate write-offs for prior four quarters; average delinquency
rate of commercial real estate for prior four
quarters . In this specification, the latter three
macro-economic variables are available on quarterly basis
only, therefore we include time fixed effects that correspond to
different origination years. The estimated
coefficient for cumulative stock returns are statistically
significant and slightly more negative than in the
regression that includes monthly time dummies. Regression
results also reveal that mortgage spreads overall
are negatively related to 10-Year maturity Treasury rate which
is consistent with theoretical predictions.
Mortgage spreads decline with the slope of the Treasury curve,
and increase with spread of AAA-rated
corporate bonds. Also, the mortgage spreads are wider after
periods when in a preceding four quarters there
is a higher delinquency rate of commercial real estate and when
the commercial real estate performs poorly
as measured by the cumulative return of NCREIF index. The
results of this regression are available upon
request.
11
-
relative to the periods when originators perform well, then the
regression with monthly
dummies will likely understate the total effect of stock price
performance on credit spreads.
As we will show in the next section, the economic impact of past
stock performance is
stronger for periods when financial industry is suffering a
downturn.
We should stress that the results should be interpreted as
cross-sectional results that
institutions with at least 15% declines in stock price originate
riskier mortgages relative to
better performing institutions during the same month period. An
alternative interpretation
of the regression results is that poor stock price performance
of the institutions is simply
driven by expectations of a decline in market fundamentals and
an increase in risk for
commercial properties, which, in turn, may explain why spreads
are wider after periods
with stock price declines. This interpretation of our regression
results is less plausible for
two reasons. First, we control for monthly fixed effects.
Second, in our data set, most
originators are large financial institutions for which
origination of commercial mortgages is
only a relatively small part of their businesses, therefor we
don’t expect that their stock
prices should be significantly driven by changes in real estate
fundamentals.
4.2.2 Estimates of Other Control Variables
The estimated coefficient of the property control variables are
consistent with our expecta-
tions. In particular, we find that mortgages on riskier property
types have larger spreads,
e.g., multi-family apartment complexes and mobile home parks
have the smallest spreads and
hotels and medical offices the largest. In addition, the
coefficient of the NOI/Property Value
ratio is significantly positive, which is consistent with theory
that suggests that a higher
expected growth rate in operating income leads to narrower
spreads. We also find that the
coefficient of the logarithm of property value is significant
and negative, which is consistent
with larger properties being less risky as well as with
economies of scale in origination and
default costs.
In contrast, the estimated coefficients for the mortgage
characteristics are somewhat
inconsistent with theory. For example, the estimated
coefficients of the LTV variable are weak
and statistically insignificant in the regression with
macro-economic variables. Similarly, the
coefficients of the loan amortization rate are small. It is
likely that the weak relation between
spreads and mortgage characteristics is due to the fact that
these are endogenous variables; to
obtain mortgages on riskier properties, originators are likely
to require lower LTV and shorter
12
-
duration. This, of course, can create a bias in our estimates of
the mortgage characteristics
coefficients, but more importantly, it can create a bias in the
coefficient of past stock returns if
there is a correlation between mortgage characteristics and past
stock returns. For example,
if mortgages originated by institutions with relatively worse
stock returns have high LTV
ratios, the coefficient on the past stock returns will be biased
upwards, because the LTV
coefficient does not appropriately capture the effect of LTV on
spreads. In Section 4.5 we
examine how past stock returns influence mortgage LTV choice to
help us gauge the extent
to which this is a problem.
4.3 Are Mortgage Spreads Wider for Worse Performing Origina-
tors During Industry Downturns?
The specification of our previous regression assumes that the
cross-sectional relation between
mortgage spreads and the fact that the mortgage is originated by
a "stock price loser" are the
same in each month. In this section we explore the possibility
that the effect is stronger in
those months in which the entire financial industry is doing
relatively poorly. Our intuition
is that the incentive to originate higher risk mortgages arises
because of pressures on the
originator when it is doing poorly, and when the potential
scrutiny of origination quality
is reduced, which means that we are more likely to see stronger
results in months in which
financial institutions, as a group, are doing quite poorly. We
expect that during these periods
an overall scrutiny of the origination quality is weaker which
would allow a "stock price loser"
to temporarily shift its origination standards.
To examine this possibility we identify periods in which the
financial industry experi-
enced economic distress and investigate whether the relation
between mortgage spreads and
originator performance is stronger in those periods. We identify
distress periods based on
the return of the S&P1500 Diversified Financials Price Index
(S&P1500 Div), which tracks
the performance of a broad range of US financial institutions.
Specifically, in the last column
of Table 4 we report the cumulative return for the prior 3
months of S&P1500 Div index
for our sample period of 1996-2002. We select the negative
cumulative stock return over
3-month period, in order to capture longer-term downward periods
in the index. During
1996-2002, the cumulative 3-month return of the index is
negative for 27 months out of 84
months, and there were 5118 mortgages in the data set originated
during these months. We
13
-
view these months as periods when financial industry suffered a
downturn. Interestingly,
these 27 months with negative index return appear to occur in
clusters and tend to overlap
with period generally known as Russian default and LTCM crises
in late 1998 as well as
a period of 9/11 aftermath and Enron collapse, but not with the
period of Asian currency
crisis. In Table 4 These three periods are identified by "*" and
"**" and "***"in the table,
respectively.
To test the hypothesis that the difference in spreads between
worse performing insti-
tutions and better performing institutions widens during
distress periods we introduce a
market downturn dummy that equals 1 if the cumulative return for
the prior 3 months of
S&P1500 Div index is negative, and zero otherwise. We then
split the "stock price loser"
dummy into two dummy variables: for periods when market downturn
dummy=1, and when
market downturn dummy=0. These two variables allow us to
disentangle the effect of "stock
price losses" on mortgage spreads in periods of industry
downturns.
Spread = intercept
+α(dummy=1 if originator is "stock price loser" and market
downturn dummy=1)
+θ(dummy=1 if originator is "stock price loser" and market
downturn dummy=0)
+ βi(property characteristics variables)
+X
γi(mortgage characteristics variables)
+ (property type dummy variables)
+ (originator dummy variable)
+ (origination time dummy variables) + .
(2)
Table 3, which reports these regression estimates, reveals that
the both coefficients for "Stock
Price Loser" dummies are positive and statistically significant.
The estimated values are
higher for periods of market downturns, which indicates that an
institution’s past stock
losses influences the spreads on the mortgages it initiates more
when financial institutions
in general are doing poorly. For example, the difference in
mortgage spreads originated by
institution whose stock return is less than negative 15% for
past 6 month is 13 b.p. This
implies that it is during times when the entire lending industry
suffer, a "stock price loser"
originates more risky mortgages when compared to a "stock price
loser" during the times
14
-
when the industry does not experience poor performance. In other
words, loosing 15%
or more during market downturns results in a larger incremental
spread relative to non-
losers if compared with losing 15% or more during normal
periods. Perhaps, during market
downturns the competition between lenders weakens leading to
even lesser quality of the
origination process and due diligence.
4.4 Cross-Sectional Estimates in Each Month
The regressions in the previous sections assumed that the
residuals were independently
distributed. Although the regressions include monthly time
dummies as well as originator
dummies, it is likely that there are sources of correlation
between the residuals that are
not adequately accounted for by our controls. For example, the
residuals of the mortgages
originated by the same institution in the same period may be
correlated. To address this
issue, we report estimates of regressions that are identical to
the regressions reported in the
last section except that they do not include originator fixed
effects and are run in each of
the individual sample month.22 By running these regressions in
individual months we can
gauge the level of significance by the extent to which the
cross-sectional relation between
the stock returns of the originator and spreads holds across
monthly periods.
Table 4 reports these regressions for each of the 84 months
between 1996-2002. The
average coefficient estimate of the "stock price loser" dummy
variable varies between 0.08
and 0.11 indicating that the credit spreads are 8-11 b.p. wider
for mortgages originated by
poorly performing institutions. For look-back window of 6
months, the estimated coefficient
for the dummy variable takes positive values for 44 months out
of 60 (for 14 months there are
no "stock price losers"). The Fama-MacBeth t-statistics is 2.77
and 3.15 implying significance
at 1% level for these two specifications. This monthly analysis
supports the hypothesis that
worst performing institutions originate riskier mortgages.
4.5 Do Poorly Performing Institutions Originate Mortgages
With
Higher LTV Ratios?
In this section we examine whether the mortgages originated by
stock price losers have
different loan to value ratios (LTVs) than those originated by
stock price winners. We
22Alternatively, we can replicate the regression on full sample
with a control for error clustering.
15
-
examine this issue for two reasons: First, as we mentioned in
the introduction, we are
concerned about the possibility that originators perform poorly
because of shocks that make
their clients riskier. If this were the case, we would expect
the poorly performing originators
to require higher LTVs on the mortgages they originate. The
second concern is that since
LTV is endogenous, the estimates of its coefficients are biased
towards zero, which means
that the regressions do not adequately control for the effect of
this characteristic. As a
result, it is possible that the wider spreads of mortgages
originated by stock price losers
arise because mortgages that are originated by these
institutions have higher LTV ratios.
To address these issues, we examine whether the relation between
LTV and the originators
past stock returns by running the following regression:
LTV = intercept+ α(dummy=1 if originator is "stock price
loser")
+X
βi(property characteristics variables)
+ (property type dummy variables)
+ (originator dummy variable)
+ (origination time dummy variables) + .
(3)
As we show in Table 5, which reports this regression, the
coefficients for the dummy
variables for stock price losers are not significant. In
addition, unreported results indicate
that poorly performing originators do not issue mortgages with
longer maturities or shift
towards risky property types (such as hotels). These results
indicate that our result that
poorly performing institutions originate mortgages with wider
spreads is not likely to be due
to biased estimates of endogenous variables that describe
mortgage terms. In addition, given
that LTV ratios are not related to the originators’ past stock
returns, we think it is unlikely
the risks of the originated mortgages are related to the
originators’ past stock returns.
5 Robustness Checks
In order to verify the robustness of the main result we redefine
the indicator for poorly
performing originator in two different ways and reestimate the
spread regression (Table 2).
First, we assume that the originator is a "stock price loser" if
its stock return declines by
at least 20% during the 3 and 6 months prior to the mortgage
origination. Compared to
16
-
the base case with the negative 15% cutoff, the fraction of
mortgages that are originated
by institutions that lost at least 20% of stock value declines.
In our data set, between 8%
and 9% of the mortgages are originated by institutions that lost
at least 20% of their stock
values in the prior one or two quarters.
The results of the spread regressions that include these new
dummy variables are reported
in Table 6. Similar to the results in Table 2, the stock price
dummy variable is positive and
significant. For example, the coefficient of the "stock price
loser" in the prior 6 months has
an estimated value of 10 b.p.
6 Alternative Explanations and Additional Tests
The evidence in previous sections indicate that originators that
experienced poor stock re-
turns in prior quarters tend to originate mortgages with wider
spreads. However, as we
discussed in the introduction, there are a number of potential
explanations for this finding.
The explanation that is our focus is that originators who are
doing poorly have an incentive
to expend fewer resources on their conduit mortgages, and are
thus likely to originate riskier
mortgages. Alternatively, the higher spreads may be due to
poorly performing financial
institutions exploiting their borrowers by charging higher
interest rates. Another possibility
is that the negative correlation between originator stock
returns and credit spreads arise
because originator stock prices perform poorly when the market
anticipates that real estate
markets become riskier. Since we have monthly time effects that
control for changes in mar-
ket wide risk, this hypothesis is not likely to be applicable to
large institutions that originate
mortgages to all segments of the market, but it could apply to
originators which tend to
focus in either certain geographical regions or on particular
property types. For example, if
an originator specializes in originating mortgages in
California, its stock price will decline
and the mortgages it originates will have higher spreads when
California real estate becomes
riskier.
In this section we present a variety of additional tests that
are designed to help us dis-
tinguish between these competing hypotheses. Most of our tests
are designed to determine
whether stock price losers originate riskier mortgages. While
these tests allow us to distin-
guish the weak underwriting hypothesis and the hypothesis that
stock price losers exploit
their borrowers, evidence of higher risk is also consistent with
the idea that stock prices fall
17
-
in anticipation of higher risk. To distinguish between the weak
underwriting hypothesis and
the later hypothesis we provide indirect evidence of sloppier
underwriting and less oversight
by originators when they are stock price losers.
6.1 Are the Results Stronger for Underwriters of CMBS Deals?
In our data set, 6,330 mortgages out of 18,526 were originated
by the same institution that
issued the CMBS that included the mortgages. When this is the
case, there is no third entity
scrutinizing the mortgage before it is ultimately sold as part
of a CMBS to investors. To
test whether the originator has a greater incentive to include
higher yielding mortgages into
CMBS deals they underwrite, we introduce an "underwriter" dummy,
which takes a value of
1 for mortgages originated by the institution that later becomes
the lead underwriter of the
CMBS in which the mortgage is sold. Our reputation hypothesis
suggests that this added
scrutiny will influence mortgage spreads during periods in which
the originator experiences
poor stock price performance. In contrast, the competing
hypotheses make no predictions
about the relevance of third party scrutiny.
The second column in Table 7 reports regression results that
indicate that uncondi-
tionally, mortgages that are originated and issued in CMBS pools
by the same institution
have about 2 b.p. higher spreads. To test whether this
difference increases when the orig-
inator/issuer has poor stock performance relative to other prior
to CMBS origination we
split the "stock price loser" dummy variable into two dummy
variables: "stock price loser—
underwriter" and "stock price loser — not underwriter." As we
report in Table 7, both
variables are positive and statistically significant for most of
the specifications, but the effect
is stronger when the stock price loser is also the CMBS
underwriter. 23 For example, the
estimated coefficients for the "stock price loser—underwriter"
dummy variables for the look-
back window of 3 and 6 months suggests that CMBS underwriters
that lost at least 15% of
their stock prices include mortgages that have spreads 10-13
b.p. wider.
23We also run similar regressions that include fixed effects for
the CMBS deal. These unreported regressions
indicate that when the CMBS underwriter is doing poorly, its own
mortgages that are included in the deals
have spreads between 10-20 b.p. wider than other mortgages in
the pool. Conversely, when the underwriter
is not doing poorly, its own mortgages included in the pool have
yields about 10-19 b.p. lower than other
mortgages in the same pool.
18
-
6.2 The Time Lag Between Mortgage Origination and CMBS Is-
suance
Originators sometimes refer to what they call "shelf risk,"
which is the risk that an event
will take place that lowers the value of a mortgage before the
mortgage is packaged and
sold to the ultimate investors. If mortgages initiated by a
stock price loser have higher
unobserved risk attributes, we would expect them to have greater
shelf risk, which should in
turn give the originator a greater incentive to go through the
packaging and issuing process
as quickly as possible.24 Hence, an analysis of the time span
between mortgage origination
and CMBS issuance can provide further indirect evidence that
stock price losers originate
riskier mortgages.
To analyze whether the originators with poor past performance
tend to sell their mort-
gages faster, we measure the time span between when the mortgage
is originated and when
it is included in a CMBS issue.25 We call this variable the
"time lag,"which is measured as
the time between the cutoff date of the deal and the mortgage
origination date. The cutoff
date is the date when the composition of the CMBS deal is first
fixed, typically the first day
of the month in which the deal is sold.26 For our mortgages, the
average “time lag” is 155
days with a standard deviation of 132 days. In the regression
the time lag is measured in
days. To capture a possible relationship between the
originators’ poor performance and the
24Another possibility is that poorly performing originators are
less able to take risk or may want to sell
the mortgages earlier to raise cash.25Due to the data
limitations stemming from the fact that for some mortgages we don’t
have information
about which CMBS pool the mortgage is sold to, our sample size
is somewhat smaller.26Sometimes new loans are added after the
cutoff date, which can result in a negative time lag We deleted
small number of mortgages with a negative time lag along with a
small number of mortgages that have "time
lags" longer than 1000 days from this regression.
19
-
time lag, we run the following regression:
Time Lag = intercept+ s(Spread)
+ α(dummy=1 is originator is "stock loser" )
+X
βi(property characteristics variables)
+X
γi(mortgage characteristics variables)
+ (property type dummy variables)
+ (originator dummy variable)
+ f(CMBS deal dummy) + .
(4)
The right hand side variables in this regression are the same as
in previous regressions except
that in addition to the property type and originator dummies,
this regression includes CMBS
deal dummies that control for the possibility that some CMBS
deals are slower to come to
market. In addition, we include mortgage spread as an
explanatory variable to control for
the possibility that higher risk mortgages are sold off more
quickly. As one can see in Table
8, the mortgages originated by “stock price losers” tend to be
sold in CMBS deals about
27-36 days quicker than other mortgages in the same deal.
This evidence is consistent with the idea that originators with
lower stock returns are
more concerned about the shelf risk of their mortgages, perhaps,
because they are concerned
about unfavorable information being revealed prior to the sale.
An alternative interpretation
is that stock price losers sell their mortgages to CMBS deals
faster because they need to raise
cash faster.27 However, our estimates also indicates that
mortgages with higher spreads are
sold to CMBS deals faster than mortgages with lower spreads.
Moreover, larger mortgages
and mortgages with higher LTV ratios tend to be securitized
faster, while mortgages with
higher loan amortization rates and mortgages with higher
NOI/(Property Value) ratios tend
to be held by institutions longer before being securitized.
These findings tend to support
the idea that originators sell mortgages more quickly when they
are viewed as riskier.
27Unreported tests indicate that, in general, the originators
that are CMBS underwriters do not sell their
mortgage to the CMBS deals faster. In addition, there is no
evidence that "stock price losers" who are also
underwriters have shorter time lags relative to other
"originator-losers".
20
-
6.3 Are Mortgages Originated by Stock Price Losers More
Likely
to Subsequently Default?
According to reputation hypothesis if originators with poor past
stock returns originate
riskier mortgages, then the mortgages are likely to default more
often, even after controlling
for mortgage and property characteristics. This will be the case
when the higher risk is
because of poor underwriting as well as because the poor stock
returns predict weak funda-
mentals. Ideally to examine default probability, we would like
to have data on the mortgage
histories up until their maturities. Unfortunately, most of the
mortgages in our data set
have not matured, so the complete default analysis is not
possible. However, we have infor-
mation about whether mortgages in the data set are marked as
"performing", "REO",28 "in
foreclosure" and "delinquent" at the time the data set was
created.29
There are a total of 158 mortgages that are marked as either "in
foreclosure" or as ROE,
and at least one month in delinquency. There are only 25
mortgages that are already in ROE
or in foreclosure. While there are differences between these
three types, given our limited
data we view them all as defaulted mortgages and define a Dummy
variable "Defaulted,"
which equals one if the status of the mortgage is either "in
foreclosure", "ROE", or the
mortgage is at least one month in delinquency. Using Defaulted
as the dependent variable,
we estimate the following probit regressions:
Probit "Defaulted" = intercept+ α(dummy=1 if originator is
"stock price loser")
+X
βi(property characteristics variables)
+X
γi(mortgage characteristics variables)
+ (property type dummy variables)
+ (origination year dummy variable) + .
(5)
Table 9, which report the results of this regression, reveals a
positive relationship (at least
at the 10% significance level) between whether the originator is
a "stock price loser" and
whether the mortgage defaults.30 These results hold even when we
include the spread as
28REO is the property acquired by the Servicer on behalf of the
Trust through Foreclosure or deed-in-lieu
of foreclosure on a defaulted loan.29Archer, Elmer, Harrison,
and, Ling (2002) classify mortgages as having defaulted if they
were late by at
least 90 days.30We, however, have to interpret this result with
some caution because our data includes only 158 defaults
21
-
an explanatory variable, which suggests that the originator’s
past stock returns capture
"unobserved risk factors" that are not incorporated into the
mortgage spreads. If poorly
performing institutions simply exploit their borrowers by
charging higher borrowing rates,
in such case we should not expect an increase in default
probabilities for those loans. A
negative relationship between stock returns and default
probability is not supportive of this
hypothesis, but rather suggests that mortgages originated by
institutions with poor stock
performance are riskier which is supportive of the reputation
hypothesis.
6.4 The Subordination Levels of the CMBS Deals
The final question we address is the extent to which credit
agencies account for the credibility
of the originator when they rate CMBS deals. To do this we
examine the AAA subordination
levels of the CMBS deals, which is the percentage value of the
total CMBS deal that are
in securities rated below AAA. For example, as reported in Table
10, in our data set, the
average AAA subordination level is 27%, which means that, if $1
Billion in CMBS securities
are issued in a deal, $730 million are senior securities that
are rated AAA, and $270 million
are junior in priority to the AAA securities and have lower
ratings. Clearly, the default risk
of the AAA securities are lower, ceteris paribus, if the
subordination level is higher. Hence,
if the rating agencies believe the mortgages in a pool are
riskier, they will require a higher
AAA subordination level. This test would allow us to distinguish
between the hypotheses
of whether poorly performing originators originate riskier
mortgages or simply overcharge
borrowers? In a latter case we should not observe an increase in
subordination level.
Our data set, which contains information about the AAA
subordination levels of 94
CMBS deals, allows us to test the extent to which the ratings
agencies account for the credi-
bility of the originator when determining the subordination
level. Specifically, we regress the
AAA-subordination level on various proxies that include the
weighted-average characteris-
tics of the mortgages included in the CMBS pool, along with the
variable that calculates
weighted-average fraction of stock price losers that sell
mortgages to the deal where weights
out of 18,526 observations.
22
-
are relative sizes of mortgages in the deal
Deal’s AAA-Subordination =
intercept+ θ(Weighted-Average Mortgage Spread )
+ α(Weighted-Average Fraction of Originators that are "Stock
Price Losers" )
+ β(Weighted-Average LTV)
+ γX(Weighted-Average Fraction of five Property Types in the
Deal)
+ φ(Deal Size)
+ ϕ(Number of Mortgages in the Deal)
+ (Deal Origination Year dummy) + .
(6)
Table 11 reports the results of this regression. The coefficient
estimates for the weighted-
average fraction of originators that are "stock price losers" in
the pool are positive and
statistically significant at the 5% level. These tests suggest
that rating agencies require
higher levels of subordination for CMBS deals that include more
mortgages originated by
underperforming originators. This finding is consistent with
reputation hypothesis and not
consistent with "overcharge" hypothesis.
It should also be noted that the estimates for year dummies
decline over time, reflecting a
general decline in subordination levels of CMBS deals since
1996.31 Also, the results suggest
that credit ratings increase with the number of mortgages
included in the deal and declines
with the weighted-average LTV. Interestingly, the weighted
average spreads of the mortgages
do not significantly affect subordination levels.
7 Conclusion
This paper provides indirect evidence of incentive/information
problems that can arise when
mortgages are originated with the intention of selling them to
investors as part of CMBS
pools. Specifically, we find that the credit spreads (i.e., the
spread between the mortgage
rate and a Treasury with the same maturity) of mortgages that
are included in CMBSs are
31Riddiough (2002) suggests that the higher subordination levels
in the earlier years was due to an initial
lack of familiarity with CMBS products.
23
-
larger when the mortgage originator experiences poor stock price
performance in the recent
past. Moreover, mortgages that are originated in these
situations are more likely to default.
This evidence, which indicates that institutions originate
riskier mortgages when they are
doing poorly, is consistent with reputation models that suggest
that firms that are doing
poorly expend fewer resources on product quality.
The issues that we address are likely to be equally applicable
to commercial loans, which
are also bundled and sold as securities. However, identifying
the effects documented in
this paper might be more challenging in a study of commercial
loans, since corporate loan
contracts tend to be somewhat more complicated and less
standardized than commercial
mortgages. Nevertheless, as we mentioned in the introduction,
Hubbard, Kuttner and Palia
(2002) consider the relation between yield spreads and the
financial condition of the lender
and also find that spreads are larger when the financial
condition of the lender is worse.
However, they suggest that the larger spreads arise because
lenders with low capitalization
ratios expropriate their borrowers. In other words, rather than
exploiting their reputation
with investors, the banks exploit their relation with their
clients.
Its likely that the channel suggested by Hubbard, Kuttner and
Palia (2002) is more
applicable in their setting, since the corporate loans in their
sample are held by the bank,
while the channel suggested by us is more applicable in our
setting, since relationships
between borrowers and originators matter very little for
mortgages that ultimately become
part of CMBS pools. However, it is possible that the incentive
to initiate riskier commercial
loans when a bank’s financial condition deteriorates is also
likely to arise even when the
loans are not sold (because there is still a trade-off between
higher profits today versus the
risk of default and lower profits in the future). Perhaps,
future research can separately
examine loans that are sold and loans that are held by the
originating bank to more directly
examine how the origination process is influenced by the
separation between the investors
and originators of debt.
References
[1] Allen, Franklin, 1984, Reputation and Product Quality, The
RAND Journal of Eco-
nomics 15 (3), pp. 311-327.
24
-
[2] Allen, Marcus T., Ronald C. Rutherford, and Marilyn K.
Wiley, 1999, The Relation-
ships Between Mortgage Rates and Capital-Market Rates Under
Alternative Market
Conditions, The Journal of Real Estate Finance and Economics 19,
211—221.
[3] Ambrose, Brent, and Anthony B. Sanders, 2003, Commercial
Mortgage-backed Securi-
ties: Prepayment and Default, The Journal of Real Estate Finance
and Economics 26,
175—192.
[4] Ambrose, Brent W., John D. Benjamin, and Peter Chinloy,
2003, Bank and Nonbank
Lenders and the Commercial Mortgage Market, The Journal of Real
Estate Finance
and Economics 26, 81—94.
[5] Archer, Wayne R., Peter J. Elmer, David M. Harrison, and
David C. Ling, 2002, De-
terminants of Multifamily Mortgage Default, Real Estate
Economics 30, 445—473.
[6] R.P. Beatty and J.R. Ritter, 1986, Investment banking,
reputation, and the under-
pricing of initial public offerings, Journal of Financial
Economics 15, pp. 213—232.
[7] Benmelech, Efraim, Mark J. Garmaise, and Tobias J.
Moskowitz, 2005, Do Liquidation
Values Affect Financial Contracts? Evidence from Commercial Loan
Contracts and
Zoning Regulation, The Quarterly Journal of Economics 120, pp
1121-1154.
[8] Carter, R.B., and Manaster, S., 1990, Initial Public
Offerings and Underwriter Reputa-
tion." The Journal of Finance 45 (4), pp. 1045-1068.
[9] Chemmanur, Thomas J., and Paolo Fulghieri, 1994, Investment
bank reputation, infor-
mation production, and financial intermediation, Journal of
Finance 49, 57—79.
[10] Fabozzi, Frank J., and David P. Jacob, 1998, The Handbook
of Commercial Mortgage-
Backed Securities, published by Frank Fabozzi Associates.
[11] Fabozzi, Frank J, Chuck Ramsey, and Michael Marz, 2000, The
handbook of nonagency
mortgage-backed securities. Publisher: New Hope, Pa.
[12] Fang, Lily Hua, 2005, Investment Bank Reputation and the
Price and Quality of Un-
derwriting Services, The Journal of Finance 60 (6),
2729—2761.
25
-
[13] Hubbard, Glenn R., Kenneth N. Kuttner, and Darius N. Palia,
2002, Are there "bank
effects" in borrowers’ costs of funds? Evidence from a matched
sample of borrowers and
banks, Journal of Business, vol. 75, no. 4, pp 559-581.
[14] Khanna, Naveen, Thomas H. Noe, and, Ramana Sonti, 2006,
Good IPOs draw in bad:
Inelastic banking capacity and hot markets, working paper.
[15] Kim, J., K. Ramaswamy and S. Sundaresan, 1993. Does default
risk in coupons affect
the valuation of corporate bonds?: A contingent claims model,
Financial Management
22(3), 117-131.
[16] Klein, Benjamin, and Keith B. Leffler, 1981, The Role of
Market Forces in Assuring
Contractual Performance, The Journal of Political Economy 89,
615-641.
[17] Livingston, Miles, and Robert E. Miller, 2000, Investment
Bank Reputation and the
Underwriting of Nonconvertible Debt, Financial Management, Vol.
29, No. 2, Summer
2000.
[18] Maris, B.A. and W. Segal. 2002. Analysis of Yield Spreads
on Commercial Mortgage-
Backed Securities. Journal of Real Estate Research 23:
235—252.
[19] Maksimovic, V. and S. Titman, 1991, Financial Policy and
Reputation for Product
Quality”, Review of Financial Studies vol 4, n1: 175-200.
[20] Michaely, Roni, and, Wayne H. Shaw, 1994, The Pricing of
Initial Public Offerings:
Tests of Adverse-Selection and Signaling Theories, The Review of
Financial Studies
7-2, pp. 279-319.
[21] Pennacchi, George, 1988, Loan Sales and the Cost of Bank
Capital, The Journal of
Finance, Vol. 43, No. 2, pp. 375-396.
[22] Riddiough, Timothy J., 1997, Optimal Design and Governance
of Asset-Backed Securi-
ties, Journal of Financial Intermediation 6, 121-152.
[23] Riddiough, Timothy J., 2002, Forces Changing Real Estate
for at least a Little While:
Market Structure and Growth Prospects of the Conduit-CMBS
Market, working paper.
26
-
[24] Riddiough, T. J., and C. Polleys, 1999, Commercial
Mortgage-Backed Securities: An
Exploration into Innovation, Product Design and Learning in
Financial Markets, Work-
ing paper.
[25] Shapiro, Carl, 1983, Premiums for High Quality Products as
Returns to Reputations,
The Quarterly Journal of Economics 98-4, pp. 659-680
[26] Titman, Sheridan, Stathis Tompaidis, and Sergey Tsyplakov,
2004, Market Imperfec-
tions, Investment Flexibility and Default Spreads, The Journal
of Finance 59, 165—206.
[27] Titman, Sheridan., and B. Trueman, 1986, Information
quality and the valuation o f
new issues, Journal of Accounting and Economics 8, 159-172.
[28] Titman, Sheridan, Tompaidis, Stathis, and Sergey Tsyplakov,
Determinants of Credit
Spreads in Commercial Mortgages, 2005, Real Estate Economics,
Volume 33, Number
4, December 2005, pp. 711-738(28).
[29] Titman, S. and W.N. Torous. 1989. Valuing Commercial
Mortgages: An Empirical
Investigation of the Contingent-Claim Approach to Pricing Risky
Debt. The Journal of
Finance 44: 345—373.
27
-
Table 1
Summary Statistics of the Commercial Mortgages in the Data
Set
Panel 1: Distribution of Commercial Mortgage Originations Across
Time Periods Year of Origination # of Mortgages
1992 20 1993 72 1994 199 1995 409 1996 1082 1997 2844 1998 5558
1999 3764 2000 1970 2001 1988 2002 1320
Panel 2: Distribution Across Property Types for Mortgages
Originated in 1996-2002
Property Type # of Mortgages Multi-family Properties 5324 Retail
Unanchored 3902 Retail Anchored 2544 Medical Office 250 Industrial
1307 Warehouse 122 Mobile Home Park 500 Office 2464 Limited Service
Hotel 843 Full Service Hospitality 197 Self Storage 595 Mixed Use
and Other Types 478
-
Table 1 –Continued
Panel 3: Mortgage and Property Characteristics at Origination
Date for Mortgages Originated in 1996-2002 This table reports the
mortgage rate spread over Treasury rate (i.e., the difference
between the rate on the mortgage and the rate on Treasury bonds
with the same maturity as the mortgage) observed on the mortgage
origination date. The loan amortization rate is defined as
1-(Principal Value at Maturity)/(Initial Principal Value).
NOI/Value ratio is the ratio of Net Operating Income divided by the
property value at origination date. Loan-to-Value ratio (LTV) is
the ratio of the face value of debt divided by the property value
at the origination date. Mean St. Dev Minimum Maximum
Spread over Treasury,% 2.23 0.73 0.19 6.14 Property Value in $M
8.19 17.59 0.05 695.00 Loan to Value (LTV) 0.70 0.11 0.04 1.00
NOI/(Property Value) 0.10 0.03 0.00 0.58
Loan Amortization Rate 0.29 0.31 0.00 1.00 Mortgage Maturity,
years 11.29 3.63 0.54 30.08
-
Table 1 –Continued Panel 4. Top 30 Originators for Mortgages
Originated in 1996-2002 This table reports the top 30 Originators
along with the characteristics of the mortgages they originate.
Name of the Originator
# of Mortgages
Spread, %
LTV
Delinquency rate
Percent of the originated mortgages included in CMBS deals that
the originator underwrites
Bank of America 1869 1.90 0.70 0.54% 88% First Union 1631 2.17
0.70 2.15% 9% FFCA 1411 3.71 0.76 0.71% 0% Lehman Brothers 1379
1.97 0.71 0.58% 100% GE Capital 1300 2.15 0.72 0.23% 0% GMAC 1030
1.95 0.70 0.19% 0% Merrill Lynch 1029 2.05 0.70 1.75% 83% Chase
Manhattan 907 2.12 0.70 0.23% 10% Bear Stearns 774 2.05 0.62 0.52%
70% Wells Fargo 554 2.17 0.59 0.18% 0% Salomon 610 2.14 0.71 0.33%
83% Morgan Stanley 590 2.05 0.68 0.54% 71% Citicorp 518 1.63 0.71
0.97% 24% Morgan Guaranty Trust 501 2.06 0.69 0.80% 53% Captec 438
3.61 0.56 7.08% 0% National Realty Funding LC 425 2.07 0.68 0.47%
0% ContiFinancial 404 2.34 0.68 0.99% 0% KeyBank 389 2.33 0.71
0.26% 0% German American Capital 353 2.00 0.71 0.00% 0% Wachovia
Bank 258 2.25 0.70 0.39% 0% Paine Webber 274 2.24 0.72 0.00% 50%
Amresco 231 2.02 0.70 1.30% 0% Goldman Sachs 280 2.15 0.73 0.71%
40% PNC 193 2.28 0.75 1.04% 0% Midland 170 2.34 0.66 1.03% 0% Impac
163 2.38 0.69 0.61% 0% CRIIMI MAE 125 1.52 0.67 0.00% 0% Credit
Suisse 105 2.44 0.70 0.91% 100% Banc One 74 2.18 0.69 1.35% 0%
Prime Capital Funding 68 1.88 0.64 0.00% 0% 20 other originators 25
1.98 0.67 0.00% 0%
-
Table 1 --Continued Panel 5. Stock Returns of Mortgage
Originators This table reports average cumulative returns of the
originators for the look-back window of 3, 6, 9 and 12 months prior
to mortgage origination. For example, the period of (-12mo, 0)
means the period of 12 calendar months prior to mortgage
origination date, where 0 is the mortgage origination date.
Prior period of
Average Cumulative returns
Standard Deviation
(-3mo, 0) 4.2% 17% (-6mo, 0) 8.32% 22%
Table 1 --Continued Panel 6. “Stock Price” Losers This table
reports fraction of mortgages originated by institutions that lost
at least 15% in periods of three, six, nine and twelve months prior
to the mortgage origination dates.
Prior periods of
Fraction of mortgages originated by institutions that
lost at least 15% in prior periods
(-3mo, 0) 11% (-6mo, 0) 10%
-
Table 2 Results of Cross Sectional Regressions for Credit
Spreads of Mortgages Originated Between 1996 and 2002
The table presents the results of the following regression:
Spread=intercept +α(Dummy =1 if the originator lost at least 15% of
its stock value in the prior periods)+∑βi(property characteristics
variables)+∑γi(mortgage
characteristics variables)+(property type dummy
variables)+(originator dummy variable) +(origination time dummy
variables)+ε, where Spread is the difference between the rate on
the mortgage and the rate on Treasury bonds with the same maturity
as the mortgage, observed on the mortgage origination date, and
measured in percentage points. The loan amortization rate is
defined as 1-(Principal Value at Maturity)/(Initial Principal
Value). NOI/Value ratio is the ratio of Net Operating Income
divided by the property value at the origination date.
Loan-to-Value ratio (LTV) is the ratio of the face value of debt
divided by the property value at the origination date.
Number of Obs.=18526 Coeff t-stat Coeff t-stat Const 2.26 32.89
2.30 34.79 Dummy =1 if cumulative stock return of mortgage
originator is less than -15% for period (-6mo, 0) 0.08 7.15 is less
than -15% for period (-3mo, 0) 0.07 6.43 Mortgage and Property
Characteristics Log(Property Value in $M) -0.14 -37.79 -0.14 -37.71
LTV 0.09 2.66 0.09 2.69 NOI/Value -0.05 -0.37 -0.06 -0.42 Loan
Amortization Rate -0.11 -4.82 -0.11 -4.84 Mortgage Maturity -0.004
-2.527 -0.004 -2.556 Property Type Dummies (vs. Mixed Use Type)
Multifamily -0.35 -17.47 -0.35 -17.54 Retail Unanchored -0.03 -1.67
-0.03 -1.65 Retail Anchored -0.13 -6.31 -0.13 -6.36 Medical Office
0.16 4.98 0.16 5.07 Industrial -0.09 -3.92 -0.09 -3.94 Warehouse
-0.08 -2.26 -0.09 -2.32 Mobile Home Park -0.25 -10.18 -0.25 -10.16
Office -0.08 -3.86 -0.08 -3.84 Limited Service Hotel 0.34 14.31
0.34 14.32 Full Service Hospitality 0.35 10.78 0.35 10.87 Self
Storage 0.05 1.89 0.05 1.82 Originator Dummy Yes Yes Time Dummy
Month Month Adjusted R-squared 0.75 0.75
-
Table 3 The regression specification is similar to one in Table
2:
Spread=intercept +α(market downturn dummy=0 )( Dummy =1 if the
originator lost at least 15% of its stock value in the prior
periods)+ 2(market downturn dummy=1 )×( Dummy =1 if the originator
lost at least 15% of its stock value in the prior periods) +
∑βi(property characteristics variables)+∑γi(mortgage
characteristics variables)+(property type dummy
variables)+(originator dummy variable) +(origination time dummy
variables)+ε,
where Spread is the difference between the rate on the mortgage
and the rate on Treasury bonds with the same maturity as the
mortgage, observed on the mortgage origination date, and measured
in percentage points. t-statistics for the variables is also
reported. Market downturn dummy =1 for each month when cumulative
return for the prior 3 months of S&P1500 Diversified Financials
price index was negative, and zero otherwise. (See Table last
column in Table 4.) There are 27 months when this index experienced
negative returns. During the selected months there are 5118
mortgages were originated in the data set. The remaining
explanatory variables are the same as in the regression described
in Table 2. For brevity we don’t report the estimates for other
variables.
Number of Obs.=18526 Coeff t-stat Coeff t-stat Const 2.30 33.35
2.30 34.83 Dummy =1 if cumulative stock return of mortgage
originator is less than -15% for period (-6mo, 0) during no market
downturn 0.03 2.08 is less than -15% for period (-6mo, 0) during
market downturn 0.13 8.03 is less than -15% for period (-3mo, 0)
during no market downturn 0.05 2.88 is less than -15% for period
(-3mo, 0) during market downturn 0.09 5.88 Mortgage and Property
Characteristics Yes Yes Property Type Dummies (vs. Mixed Use Type)
Yes Yes Originator Dummy Yes Yes Time Dummy Month Month Adjusted
R-squared 0.75 0.75
-
Table 4 Estimated coefficients for the “Stock Price Loser Dummy”
variable in the Spread regression run separately for each
individual origination month
The table reports estimated coefficients for the Dummy
Variable=1 if stock return of the originator declined by at least
15% in the spread regression run separately for individual month of
mortgage origination. The regression is as follows:
Spread=intercept +α(cumulative stock return of mortgage
originator is less than 15% over look-back window)+∑βi(property
characteristics variables)+∑γi(mortgage characteristics
variables)+(property type dummy variables)+ε,
where Spread is the difference between the rate on the mortgage
and the rate on Treasury bonds with the same maturity as the
mortgage, observed on the mortgage origination date, and measured
in percentage points. The variables of property and mortgage
characteristics are the same as in the regression presented in
Table 2. “-“ denotes months in which no “stock price losers”
participated. For brevity we don’t report the estimates for the
other variables. The Fama-McBeth t-statistic is reported in the
last row.
Estimate for Cumulative For Dummy Variable=1 if stock return of
the
originator declined by at least 15% for the period of Cumulative
return for the prior 3 months of S&P1500
Diversified Financials -- price index YEAR MONTH [-6mo, 0]
[-3mo, 0]
1996 1 - -0.03 0.01 2 - 0.35 0.13 3 - - 0.11 4 - - 0.14 5 - -
0.01 6 - - -0.01 7 0.09 - -0.01 8 -0.18 -0.18 0.03 9 0.12 0.07 0.01
10 - - 0.06 11 -0.03 -0.03 0.12 12 - 0.11 0.24
1997 1 - -0.16 0.13 2 - - 0.18 3 - - 0.11 4 0.11 0.00 0.02 5
0.12 0.12 0.02 6 - - 0.05 7 - - 0.26 8 - - 0.27 9 - 1.98 0.11 10
0.48 - 0.14 11* 0.27 - 0.04 12* 0.34 0.35 0.18
1998 1* 0.14 0.29 0.09 2* 0.12 0.49 0.13 3* -0.05 - 0.12 4 -0.14
0.18 0.15 5 0.12 0.12 0.13 6 0.22 0.03 0.06 7 -0.12 0.16 0.09 8**
-0.08 -0.09 0.02 9** -0.06 -0.04 -0.17 10** 0.16 0.16 -0.27 11**
0.15 0.20 -0.08 12** -0.80 0.26 0.24
-
Table 4—continued
Estimate for Cumulative For Dummy Variable=1 if stock return of
the
originator declined by at least 15% for the period of Cumulative
return for the prior 3 months of S&P1500
Diversified Financials – price index YEAR MONTH [-6mo, 0] [-3mo,
0]
1999 1** - 0.05 0.36 2 0.17 -0.08 0.20 3 -0.05 -0.07 0.13 4 0.02
-0.20 0.16 5 - 0.06 0.20 6 0.22 -0.03 0.06 7 -0.47 -0.31 0.07 8
0.22 - -0.11 9 -0.35 -0.27 -0.04 10 0.06 0.03 -0.15 11 0.06 -0.04
0.09 12 0.31 0.24 0.10
2000 1 0.33 0.04 0.14 2 0.06 0.14 0.00 3 0.10 0.08 -0.03 4 0.32
0.25 0.15 5 0.23 0.25 0.07 6 0.06 0.34 0.14 7 -0.16 0.21 -0.02 8
-0.19 -0.12 0.13 9 - -0.22 0.20 10 0.54 0.21 0.23 11 0.03 0.02 0.07
12 -0.03 0.03 -0.10
2001 1 0.21 0.69 -0.04 2 - -0.06 0.06 3 0.12 0.01 0.01 4 0.03
0.17 -0.14 5 -0.12 0.03 -0.10 6 0.14 0.12 0.05 7 0.56 0.17 0.11 8
-0.01 -0.05 -0.01 9**** 0.04 0.05 -0.13 10**** 0.05 0.01 -0.21
11**** 0.24 0.09 -0.15 12**** - 0.08 -0.03
2002 1**** - 0.01 0.14 2**** 0.16 0.14 0.04 3**** 0.80 0.45
-0.01 4**** - - 0.01 5**** - 0.14 -0.01 6**** 0.02 0.02 -0.03 7****
0.01 0.10 -0.16 8**** 0.12 0.22 -0.21 9**** 0.04 0.19 -0.13
-
10**** 0.06 0.17 -0.16 11**** 0.13 -0.03 0.05 12**** - -0.14
0.04
Average
0.08
0.11
Fama -McBeth Statistics 2.77
3.15
-
Table 5 Loan-to-Value Regressions
The table reports the results of cross sectional regressions of
the LTV of the mortgages on mortgage and property characteristics
along with dummies for the past returns of the originator. The
regression we estimate is as follows:
LTV=intercept +α(Dummy =1 if the originator lost at least 15% of
its stock value in the prior periods) +∑βi(property characteristics
variables)+(property type dummy variables)+(originator dummy
variables)+(origination time dummy variables)+ε,
where Loan-to-Value ratio (LTV) is the ratio of the face value
of debt divided by property value at origination. NOI/Value ratio
is the ratio of Net Income divided by the property value.
Number of Obs.=18526 Coeff t-stat Coeff t-stat Const 0.69 50.59
0.69 50.94 Dummy =1 if cumulative stock return of mortgage
originator is less than -15% for period (-6mo, 0) 0.003 1.020 is
less than -15% for period (-3mo, 0) 0.001 0.281 Mortgage and
Property Characteristics log(Property Value in $M) -0.01 -6.42
-0.01 -6.41 NOI/Value 0.59 12.88 0.58 12.87 Property Type Dummies
(vs. Mixed Use Type) Multifamily 0.05 10.10 0.05 10.11 Retail
Unanchored 0.00 0.35 0.00 0.37 Retail Anchored 0.03 6.77 0.03 6.78
Medical Office -0.02 -2.26 -0.02 -2.24 Industrial 0.01 1.09 0.01
1.10 Warehouse 0.02 1.81 0.02 1.81 Mobile Home Park 0.00 0.31 0.00
0.33 Office 0.00 -0.62 0.00 -0.60 Limited Service Hotel -0.06 -9.89
-0.06 -9.88 Full Service Hospitality -0.08 -8.05 -0.08 -8.04 Self
Storage -0.03 -4.75 -0.03 -4.75 Originator Dummy Yes Yes
Time Dummy Month Month
Adjusted R-squared 0.23 0.23
-
Table 6 Robustness Check: Alternative Indicator Variables for
Stock Performance of Originator
Table reports the results of cross-sectional spread regressions
that include indicator variables for originators that experienced a
negative stock return of at least 20% prior to the mortgage
origination dates. Specifically, we estimate the following
regression: Spread=intercept + α(Dummy =1 if the originator lost at
least 20% of its stock value in the prior periods) + ∑βi(property
characteristics variables)+∑γi(mortgage characteristics
variables)+(property type dummy variables)+(originator dummy
variable) +(origination time dummy variables)+ε, where Spread is
the difference between the rate on the mortgage and the rate on
Treasury bonds with the same maturity as the mortgage, observed on
the mortgage origination date, and measured in percentage points.
The remaining explanatory variables are the same as in the
regression described in Table 2. For brevity we don’t report the
estimates for other variables. Number of Obs.=18526 Coeff t-stat
Coeff t-stat Dummy=1 if for stock returns of the mortgage
originator is less than -20% for period (-6mo, 0) 0.10 7.8 is less
than -20% for period (-3mo, 0) 0.07 4.9 Mortgage and Property
Characteristics Yes Yes Property Type Dummies (vs. Mixed Use Type)
Yes Yes Originator Dummy Yes Yes Time Dummy Month Month Adjusted
R-squared 0.75 0.75
-
Table 7
Spread Regression for Underwriters/Non-underwriters of CMBS
deals The table report the estimates of the dummy variables for
originators that are “stock price losers”. In the regression the
dummy variable is split into two dummy variables: "stock price
looser--underwriter" and "stock price looser -- not
underwriter":
Spread = intercept + a*(dummy=1 if the originator is CMBS
underwriter and its stock declined by at least 15% prior to
origination)+ b*(dummy=1 if the originator is NOT CMBS underwriter
and its stock declined by at least 15% prior to origination) +a*(
variables of property and mortgage characteristics) +f*(Dummy For
Mortgage Originator) +g*(Dummy for Mortgage Origination Year or
Quarter)+e,
where Spread is the difference between the rate on the mortgage
and the rate on Treasury bonds with the same maturity as the
mortgage, observed on the mortgage origination date, and measured
in percentage points. The "underwriter" dummy=1 for the mortgages
issued by the originator who later becomes the lead underwriter of
the CMBS in which the mortgage is sold, and zero otherwise. The
loan amortization rate is defined as 1-(Principal Value at
Maturity)/(Initial Principal Value). NOI/Value ratio is the ratio
of Net Operating Income divided by the property value at
origination date. Loan-to-Value ratio (LTV) is the ratio of the
face value of debt divided by property value at origination.
Number of Obs.=18526 Coeff t-stat Coeff t-stat Coeff t-stat
Dummy=1 if originator that is also CMBS underwriter 0.023 2.28
Dummy=1 if originator is CMBS underwriter and its cumulative stock
return is less than -15% for period (-6mo, 0) 0.13 7.12 is less
than -15% for period (-3mo, 0) 0.10 4.91 Dummy=1 if originator that
is NOT CMBS underwriter and its cumulative stock return is less
than -15% for period (-6mo, 0) 0.05 4.28 is less than -15% for
period (-3mo, 0) 0.06 4.92 Mortgage and Property Characteristics
Yes Yes Yes Property Type Dummies (vs. Mixed Use Type) Yes Yes Yes
Originator Dummy Yes Yes Yes Time Dummy Month Month Month Adjusted
R-squared 0.75 0.75 0.75
-
Table 8 “Time Lag” Regressions
This table reports the results of the following Tobit
regression: Time Lag (in Days)=intercept + s*(Spread)+α (Dummy =1
if the originator lost at least 15% of its stock value in the prior
periods)+∑βi(property characteristics
variables)+∑γi(mortgage characteristics variables)+(property
type dummy variables)+(originator dummy variable) +(origination
time dummy variables)+ε,
where the Time Lag is the time span (measured in days) between
when the mortgage is originated and when the CMBS composition is
finalized. The loan amortization rate is defined as 1-(Principal
Value at Maturity)/(Initial Principal Value). NOI/Value ratio is
the ratio of Net Operating Income divided by the property value at
origination date. Loan-to-Value ratio (LTV) is the ratio of the
face value of debt divided by property value at origination.
Dummies for each of the CMBS deals are included.
Number of Obs.=14420 Coeff t-stat Coeff t-stat Dummy=1 if
originator’s cumulative stock return is less than -15% for period
(-6mo, 0) -36.42 -9.86 is less than -15% for period (-3mo, 0) -