J Real Estate Finan Econ

DOI 10.1007/s11146-016-9548-1

From Origination to Renegotiation: A Comparison

of Portfolio and Securitized Commercial Real
Estate Loans

Lamont Black 1 & John Krainer 2 & Joseph Nichols 3

# Springer Science+Business Media New York (outside the USA) 2016

Abstract We use a unique loan-level dataset to compare portfolio and securitized

commercial real estate loans. The paper documents how the types of loans banks
choose to hold in their portfolios differ substantially from the types of loans the same
banks securitize. Banks tend to hold loans that are Bnon-standard^ in some observable
dimension. These loans are riskier and more likely to become delinquent or distressed.
Conditional on default, we find that banks are significantly more likely to extend
portfolio loans than is the case for securitized loans. Our results suggest that banks
have a comparative advantage in funding risky assets with contracts that may require
flexibility in the event of distress.

Keywords Securitization . CMBS . Commercial banks . Asymmetric information .

Renegotiation

JEL Classification G14 . G21 . D82 . G33

The views expressed are not necessarily those of the Board of Governors of the Federal Reserve System or the
Federal Reserve Bank of San Francisco.

* Joseph Nichols
[email protected]
Lamont Black
[email protected]
John Krainer
[email protected]

1
Driehaus College of Business, DePaul University, 1 E. Jackson Blvd, Chicago, IL 60604, USA
2
Federal Reserve Bank of San Francisco, P.O. Box 7702, San Francisco, CA 94120, USA
3
Board of Governors of the Federal Reserve System, 20th and C Str. NW, Washington,
D.C. 20551, USA
 L. Black et al.

Introduction

One of the most important trends in credit markets over the past 30 years has been the
steady increase in debt held by investors in a securitization structure. The commercial
real estate (CRE) asset class is no exception. Since the late 1990s, the dollar amount of
securitized CRE has increased five-fold to more than one-half trillion dollars. Yet,
despite this rapid growth rate, the share of securitized CRE is still only about 20 % of
the total CRE mortgage debt (see Fig. 1). Of the remaining 80 %, the majority of the
loans are held on the balance sheets of banking institutions.1 In this paper we explore
the characteristics of CRE loans originated by commercial banks and the determinants
of the loan securitization decision. We shed light on the differences between portfolio
loans and securitized loans observed at origination, as well as the differences in lender
behavior throughout the life of the loans.
To address these questions we use confidential supervisory loan-level data on
portfolio loans collected as part of the recent stress tests of U.S. banks mandated by
the Dodd-Frank Act. Large banks are required to provide loan-level data on their
commercial real estate loan portfolios in order to support the Comprehensive Capital
Analysis and Review (CCAR) along with other supervisory programs. These data
provide a unique opportunity to analyze portfolio loans from origination to renegoti-
ation. We observe loan origination characteristics, loan performance over time, as well
as bank efforts to renegotiate these loans in the event of distress.
We combine the stress test data with Morningstar data on loans pooled in commer-
cial mortgage-backed securities (CMBS). The CMBS data include very similar infor-
mation on loan characteristics and loan performance. We observe the types of loans in
CMBS deals and the likelihood of a loan in distress being extended by the special
servicer of the securitized loans. In addition, we can compare loans by the same
originator across the two data sources, because many of the loans in the CMBS data
were originated by the banks in our stress test data.
Based on these rich data, we document substantial differences between portfolio and
securitized commercial real estate loans. CMBS loans are almost entirely fixed-rate
loans on stabilized income-producing properties.2 These types of loans typically imply
a steady stream of predictable cash flows for capital market investors in CMBS. In
contrast, CRE loans that are held in commercial bank portfolios tend to be Bnon-
standard^ in some observable dimension. For instance, a large number of bank loans
have a floating interest rate or are secured by properties that are under construction or
owner-occupied. These types of loans do not appear in loan pools for CMBS deals.
This is the first indication that banks have a comparative advantage in funding
commercial real estate projects that are less suitable for arm’s length capital markets.
One of the factors that influences the portfolio-versus-securitization choice is the
potential of banks to renegotiate loans. This renegotiation potential could stem from a
variety of sources. Banks may have repeated interactions with the same borrower or

1
The share of securitization here includes CRE loans held by banks that are not good candidates for
securitization, such as construction loans and loans on owner-occupied properties. Later in the paper we
restrict our analysis to the pool of bank loans that are potential candidates for securitization, i.e. fixed rate loans
on income producing properties.
2
In other words, the property has tenants making regular payments to the property owner and the loan on the
property has a fixed interest rate.
From Origination to Renegotiation

Securitized Commercial Mortgages

.25
600

.2
400

.15
share
.1
200

.05
0

0
1985q1 1990q1 1995q1 2000q1 2005q1 2010q1 2015q1
dateq

Securitized comm. mortgages Percent all comm. mortgages

Source: Flow of Funds/Federal Reserve

Fig. 1 Volume and share of securitized commercial mortgages. The blue line shows the amount of commer-
cial mortgages held in asset-backed securities (ABS) in billions. The red line shows the share of all
commercial mortgages that are securitized. Total commercial mortgages here include construction loans, loans
on owner-occupied properties, and loan on income producing properties

have access to other informational advantages that allow them to better monitor loan
performance. Given the design of our study pairing portfolio and securitized loans
originated by the same banks, it may be the case that the original underwriting produces
information that is particularly useful in case of distress. Finally, there are legal
contracts that may limit a CMBS special servicer’s flexibility to resolve troubled loans.
The special servicer is the designated party for managing distressed loans in a CMBS
pool, but the special servicer’s function is narrowly prescribed by a pooling and
servicing agreement among the diverse set of CMBS investors.
All of these differences between portfolio and securitized lending combine to form
the basis of our empirical investigation. If banks have a comparative advantage in
renegotiation, then there will be segmentation in the market for CRE loans, with bank
portfolios attracting loans that benefit the most from renegotiation. This yields two
predictions that can be tested in our data: banks fund riskier loans than CMBS investors
and banks are more likely to extend loans in distress.
To test these predictions, we do a direct comparison of loan characteristics and
performance for a sample of portfolio and securitized loans that were originated by the
same group of banks. This approach is similar to Ghent and Valkanov (2013) and
Downs and Xu (2014). We restrict the sample to fixed-rate, income-producing loans, to
produce a sample of loans with overlapping characteristics in both the CRE portfolios
and CMBS deals. This narrows our sample to the loans that, based on the underlying
loan and property characteristics, could potentially have qualified for CMBS financing.
In addition, we only include loans from banks that do both portfolio and securitization
 L. Black et al.

lending. This allows us to use bank-fixed-effects to analyze the portfolio-securitization

decision within a bank over time. We study the loan originator’s actions and incentives
both at the time of origination as well as later in the loan history, with a particular
emphasis on behavior in the case of borrower distress.
Our analysis proceeds in three steps. We first look at which types of loans within our
overlapping sample are more likely to be securitized. We then look at whether portfolio
or securitized loans are more likely to go into distress. Lastly, conditional on loan
default, we analyze whether portfolio or securitized loans are more likely to be
extended.
To preview our results, even within our overlapping sample of fixed-rate loans on
income-producing properties, we find significant differences between portfolio and
securitized loans. Controlling for observable borrower and loan characteristics, we find
that portfolio loans have higher interest rates than securitized loans. While this could
reflect the higher cost of funds for banks compared to CMBS, it may also indicate
greater risk as well as a desire by the borrower to Bpurchase^ an option to renegotiate.
We find support for the second hypothesis by showing that commercial mortgages
originated by banks and held on the balance sheet are more likely to encounter some
kind of distress. In addition we finds that the original pricing of the loan is highly
predictive of subsequent distress among bank loans, but less so among CMBS loans.
Finally, conditional on default, we show that banks are much more likely to renegotiate
defaulted loans than are the special servicers and investors in CMBS. All of these
results together support a view of the market for CRE finance where lenders specialize
in different forms of intermediation.
The paper is organized as follows. Related Literature section reviews the literature
and discusses our contribution. Data and Descriptive Statistics section describes the
data and provides a number of descriptive statistics. Empirical Methodology section
outlines the methodology for the analysis. Results section presents the empirical results,
including a set of results using propensity score matching to control for endogenity.
Conclusion section concludes.

Related Literature

The role of securitization in the financial crisis has motivated a large and growing
literature on the structure and implications of securitization. Much of this literature has
focused on adverse selection and the Blemons problem,^ building on early models of
information asymmetry as in Akerlof (1970).
A number of post-financial-crisis papers explore distortions in the origination of
loans underlying residential mortgage-backed securities (RMBS). Puranandam (2011)
finds evidence that the underwriting standards for securitized residential loans is
affected by the capital structure of the originator. Keys et al. (2009) compare securitized
residential mortgages with those held on-balance-sheet and find that the securitized
mortgages are lower quality. Much of this work focuses on the incentives for loan
monitoring. Coleman et al. (2006) find that the degree of loan monitoring does indeed
affect the terms of a loan.
The literature on commercial mortgages has focused on many of the same issues as
the residential mortgage literature. Many early studies focused on the determinants of
From Origination to Renegotiation

commercial mortgage defaults using data from one or more life insurance companies
(Snyderman 1991; Esaki et al. 1999; Vandell et al. 1993; Ciochetti et al. 2003). The
literature on the CMBS market has grown along with the securitization of commercial
mortgages. A number of these papers have also focused on the defaults of loans in the
securitized pool (Ambrose and Sanders 2003; Archer et al. 2002; and Deng et al. 2004).
In addressing this issue, these papers document the correlation of loan performance
with observable underwriting characteristics such as loan-to-value (LTV) ratio and
debt-service-coverage-ratio (DSCR). An et al. (2011) analyze loans in CMBS deals
to show that loans originated by conduit lenders enjoy a 34 basis point pricing
advantage over loans originated by portfolio lenders.
Our paper is related to other recent work on securitized commercial loans exploring
issues associated with the loans in CMBS deals. For instance, Titman and Tsyplakov
(2010) and Black et al. (2012) show that originators’ financial health and organizational
form affects the quality of securitized loan underwriting. Furfine (2010) finds evidence
that the increasing complexity of CMBS deals allows originators to include lower-
quality loans in CMBS pools.
The analysis in our paper focuses less on asymmetric information and the lemons
problem, and instead builds on recent work exploring how specialization amongst
lenders might lead to differences in loan contracts and market segmentation. In
particular, our work emphasizes the role of flexibility and renegotiation in the loan
contract. Our focus on commercial real estate is most closely related to Ghent and
Valkanov (2013) and Downs and Xu (2014), who also examine the differences in
portfolio and securitized CRE loans. Using a different data set, Ghent and Valkanov
compare loan characteristics of retained and securitized loans and find that loan size
is an important predictor of securitization. Their result on loan size leads them to
conclude that diversification is likely to be a primary motive for securitization.
Ghent and Valkanov find no meaningful differences in loan performance between
securitized and portfolio loans. However, consistent with our paper, they find that
securitized loans that default take more time to resolve. Downs and Xu find that
once a loan is in distress, portfolio loans are more likely to be foreclosed upon than
are securitized loans. Like Ghent and Valkanov, Downs and Xu also find that
resolution times are slower for securitized loans than portfolio loans. Our focus
on one particular type of resolution—renegotiation and extension—is more closely
related to residential mortgage papers of Piskorski et al. (2010) and Agarwal et al.
(2011). We contribute to the literature by using unique bank portfolio data to
highlight banks’ renegotiation of distressed loans. The higher likelihood of receiv-
ing an extension when distressed may help explain why some borrowers choose
bank financing for their CRE loans, despite higher interest rates, over CMBS
financing.
Our approach to modeling differences between portfolio and securitized loans also
relates to models of investment flexibility. These models focus on the value associated
with the ability to renegotiate a financial contract. The flexibility of these contracts, in
certain cases, makes them more attractive to both sides than more rigid inflexible
contracts. Titman et al. (2005) calibrate their model of investment flexibility using data
on office buildings and commercial mortgages. In our paper, we focus on the invest-
ment flexibility of bank lenders relative to the more constrained optimization of CMBS
servicers.
 L. Black et al.

Data and Descriptive Statistics

Data on Bank Loans

As part of the Comprehensive Capital Analysis and Review (CCAR), the Federal
Reserve collects loan-level data on bank portfolios of commercial real estate
loans. 3 These data contain the most detailed information on commercial real
estate loans the Federal Reserve has ever collected.4 All bank holding companies
with $50 billion or more in consolidated assets must complete the FR Y-14
regulatory reporting form for Wholesale Risk, which includes granular data on
commercial real estate loans. As of 2014:Q3 FR Y-14 respondents held $13.9
trillion in assets, accounting for 70 % of the assets held by all BHCs, savings
and loan holding companies, and commercial and savings banks.5 The loan-level
characteristics include measures observed both at origination and throughout the
history of the loan.6 This includes origination date, loan balance, property type,
loan rate, interest rate type (fixed versus floating), maturity, loan-to-value, and
loan purpose (construction, income-producing, owner-occupied). To be included
in the data collection the loans must have a minimum size of $1 million. The
data provide a quarterly snapshot of the loan portfolio of each CCAR bank
beginning in the first quarter of 2012.7
Figure 2 shows the distribution of loan types in the portfolios of our sample of banks
in 2012:Q1. Fixed-rate loans on income producing properties account for only 13 % of
the portfolio, with construction loans, adjustable-rate loans on income-producing
properties, and owner-occupied loans all having equal or greater shares.8
Figure 3 graphs the differences in original maturity across the different types of
loans. Construction loans tend to have short, 1–3 year maturity schedules. Owner-
occupied and fixed-rate income-producing loans have similar maturity profiles.
Adjustable-rate income producing loans have a wider range of maturity terms, with
concentrations in both the shorter and the longer maturities.
Table 1 documents significant differences across bank loan types in terms of size,
underwriting, and pricing. Note, too, the broad range of bank lending that is apparent in
the FR Y-14 data. Our analysis below will focus on the risk and renegotiation properties
of the income-producing fixed-rate category, which is the smallest category in Table 1.
Since this category is also the least risky category in the bank CRE portfolios, we
believe that many of our conclusions will apply to the other loan categories as well.

3
FR Y-14Q Reporting Form and Instructions: http://www.federalreserve.gov/apps/reportforms/reportdetail.
aspx?sOoYJ+5BzDZGWnsSjRJKDwRxOb5Kb1hL
4
We expect that over time, as the collection matures, it will become an invaluable source of information
regarding the CRE market and the banks’ participation in that market.
5
Source: Call Report and Y9C.
6
As the FR Y-14 is a new data collection, we do not have reliable data for all of the fields that we require for
analysis prior to 2012:Q1. As a result we have a left-censored database. We observe the portfolio loans that are
still current as of 2012:Q1, but not those that have been originated, held in portfolio, and then resolved prior to
that date.
7
Although this misses the height of the crisis, there is still significant stress on the loans in the sample.
8
We will refer to fixed-rate loans on income producing properties as fixed-rate income producing loans for the
remainder of the paper. The income producing refers to the nature of the collateral, not the loan contract.
From Origination to Renegotiation

Construction
15%

Owner
Occupied
33%
Income
Producing -
Adjustable
39%
Income
Producing - Fixed
Rate
13%

Fig. 2 Share of bank portfolios by loan type. The pie chart shows the share of bank portfolios by loan type for
a set of banks active in origination for both securitization and their own portfolios as of 2012:Q1. The
distribution is based on the number of loans and is not weighted by outstanding balance. Source: Federal
Reserve (FR Y-14)

Data on Securitized Loans

The paper also uses CMBS data from Morningstar LLC. This compilation of loan-
level data includes every CRE loan in publicly issued (including 144 A) CMBS
deals over the period of our sample. It includes loan level credit characteristics
including the vacancy rate on the property, net operating income (NOI), loan-to-
value (LTV), and other key components. Morningstar also tracks loan perfor-
mance, with detailed information on delinquency and, to a lesser extent, loss-
given-default.

50.0%
Construction
45.0%
40.0%
Income Producing - Adjustable
35.0%
30.0% Income Producing - Fixed Rate

25.0%
Owner Occupied
20.0%
15.0%
10.0%
5.0%
0.0%
1 2 3 4 5 6 7 8 9 10 11+
Original Term In Years
Fig. 3 Distribution of bank portfolios by original term. The bar chart reports the distribution of the original
term in years for loans in the portfolios of our sample of banks in the first quarter of 2012. The blue bars report
the distribution for construction loans, the red for adjustable rate loans on income producing properties, the
green on fixed rate loans on income producing properties, and the purple for loans on owner occupied
properties. Source: Federal Reserve (FR Y-14)
 L. Black et al.

Table 1 Characteristics of bank loans by loan type. The table below reports the characteristics of bank
portfolios by loan type for a set of banks active in origination for both securitization and their own portfolios as
of 2012:Q1. The values for current balance, current LTV, and current interest rate are sample means

Construction Income Income Owner

Producing – Fixed Producing – Adjustable Occupied

Current Balance ($ millions) $ 5.6 $ 3.7 $5.3 $2.7

Current LTV 59.1 % 55.6 % 56.1 % 58.4 %
Current Interest Rate 3.7 % 5.6 % 3.8 % 4.2 %
Number of Loans 10,119 8620 26,115 21,860

Source: Federal Reserve (Y-14)

Constructing Our Sample For Comparison

We combine these data by restricting our CMBS database to those loans active as of the
first quarter of 2012, which corresponds to the sample period for which we have
available FR Y-14 bank data. 9 We also focus exclusively on the set of banks active
in originations for both securitization and portfolios. The availability of the granular
information on banks’ portfolios allows us to directly compare at the loan level
securitized and portfolio loans originated by the same institution. In other words, our
sample only includes lenders that originated CRE loans both for their loan portfolios
and for securitization. By narrowing the sample to these banks, we are able to focus on
the active decision this group of banks made about whether to securitize a loan.10
Ideally, our analysis would compare all loans in our sample of bank CRE portfolios
with all loans that ended up in CMBS. However, for reasons discussed above, there
may be a natural matching taking place between banks and borrowers that can benefit
from the banks’ information production. These differences amongst borrowers may
imply that CRE loans in bank portfolios are observably different than the CRE loans in
CMBS pools. Indeed, some of the differences that stand out are that loans in bank
portfolios tend to be of smaller size, of shorter maturity, and more likely to be floating
rate. Any one of these characteristics may cause a loan to be Bnon-standard^ in a way
that makes it difficult or impossible to securitize. This complicates the empirical
analysis. If a certain loan attribute (say floating rate interest) is overwhelmingly linked
with a certain source of loans (say, bank portfolio), then it is difficult to infer how that
loan would have performed if it was funded by another source (say, CMBS in this
case).
To combat this problem we limit our analysis to fixed-rate income-producing loans.
In other words, we focus on the set of loans that are comparable to the standard CMBS
loan.11 This highlights the Bmargin^ on which both banks and borrowers evaluate the

9
The Morningstar data extends back to the mid-1990s. FR Y-14 data available prior to 2012:Q1 lacks key
variables needed for our analysis.
10
The Appendix provides a detailed discussion about the sample construction, including a comparison of
descriptive statistics of the sample with the excluded CMBS and bank loans. This supports our contention that
the sample is representative of the share of the CRE loan market where securitization is a viable option.
11
Later in the paper we will shrink the sample even further through a propensity score matching approach to
attempt to make the loans in the sample even more comparable.
From Origination to Renegotiation

benefits/costs of the securitization decision. We will defer a deeper examination into the
other segments of CRE bank lending (i.e., construction loans and non-income-produc-
ing) for further research.
We restrict the sample to the main five property types (office, retail, industrial, hotel,
and multifamily), excluding mixed- use and other non-standard property types.
Once we account for observations that are missing key variables we have 5373 bank
loans and 19,487 CMBS loans in our analysis.

A Comparison of Similar Bank and Securitized Loans

Table 2 and Figs. 4 and 5 show simple comparisons of some of the loan and price
characteristics of CRE loans retained in bank portfolios and those sold into CMBS
pools.12 Consistent with the findings of Ghent and Valkanov (2013) and Downs and Xu
(2014), securitized loans are significantly larger. We also find that average loan to value
ratios (LTVs) at origination are slightly lower for portfolio loans (about 60 %) com-
pared to securitized loans (69 %), possibly reflecting the fact that property-level
fundamentals in the portfolio loans are more risky, therefore the lenders require more
equity buffer in exchange for financing. This hypothesis is supported when we compare
the current occupancy rates reported for each sample, with the average for properties
financed by securitized loans at 89 % and the average for those financed by bank loans
slightly lower at 84 %. Portfolio loans also require higher debt yield at origination and
wider spreads over treasuries, suggesting that banks are pricing the higher risk seen in
these loans.
When we turn to realized default rates, however, we see that the share of the
portfolio currently past due or having been previously extended as of 2012:Q1 is lower
in the bank loans (5.2 %) than what is observed in the CMBS loans (14.7 %). We
include loans that were previously extended past their original maturities in this
measure as we believe that those represent distressed loans where the extension was
part of a loss mitigation strategy adopted by the lender. The left censoring of the data
makes it difficult to interpret the meaning of this observed difference. We do not
observe any of the distressed loans that were resolved prior to this period, just those
that are still in the portfolio. This is why we limit ourselves later in the paper to an
analysis of future distress using a sample of loans that are both current and have never
been extended as of 2012:Q1, in order to better isolate the risk drivers of loans entering
distress.
Perhaps most interesting in these univariate comparisons are the distributions of loan
term in Fig. 4. There appears to be a significant clustering in the CMBS data, with
about 80 % of the CMBS data having an original loan term of 10 years. By contrast,
about 40 % of the loans in the retained portfolio have original loan terms distributed in
the 1–7 year range. This fact is consistent with our market segmentation story where
lenders for risky projects might want to shorten the term of the loan so as to speed up
the timing of the information production process.
It is well-known that property type is an important determinant of CRE loan risk.
The bar chart in Fig. 5 suggests that the property type representation is fairly similar

12
All of the loans in our sample are income-producing, which means that the initial construction phase has
been completed and the vacancy rate has stabilized to normal levels.
 L. Black et al.

Table 2 Characteristics of CMBS and bank loans. The table below compares the characteristics of CMBS
and bank loans within the subsample of fixed-rate loans on income producing properties as of 2012:Q1. Both
samples are limited to loans from the same set of banks active in origination for both securitization and their
own portfolios

CMBS Banks

Current Balance ($ millions) $ 13.8 (35.4) $ 4.4 (10.1)

Balance at Origination ($ millions) $ 15.0 (40.7) $ 4.0 (10.1)
Current LTV 62.5 % (12.8) 55.7 % (19.7)
LTV at Origination 69.1 % (11.3) 60.1 % (20.4)
Current Occupancy Rate 88.9 % (17.6) 83.7 % (28.5)
Current Debt Yield 13.0 % (9.7) 15.6 % (12.1)
Debt Yield at Origination 11.5 % (5.8) 13.6 % (10.9)
Spread to Treasuries 1.5 % (0.7) 2.4 % (1.5)
Percent in Past Due Or Previously Extended 14.7 % 5.2 %

Standard Deviations are reported in ()s

Source: Federal Reserve (Y-14)

across lender types. The main differences appear to be that large retail properties are
more likely to be funded through the CMBS market, while loans on multi-family
properties are somewhat more likely to be kept on the originator balance sheet.
Table 3 shows the percentage of CRE loans securitized by year, conditional on the
loans being active as of 2012:Q1. The first two columns report the number of loans
held in CMBS and in bank portfolios for each year from 2000 to 2012.13 These levels
indicate that the propensity for a loan to be securitized increased in the years prior to the
crisis, as shown with the aggregate data in Fig. 1. The left censoring of the data does
introduce a bias into some of these measures. If banks tend to originate loans with
shorter terms, as we documented in Fig. 5, those loans that were originated at the same
time as a longer-term CMBS loan have resolved and are not included in this analysis.
This may introduce an upward bias in our measurement of the propensity to securitize
in earlier origination years.
Table 4 compares the rate of default from 2012:Q1 to 2014:Q3 in the data for
CRE loans in CMBS and those held by banks. Of the loans securitized, 4.9 % go
into default during the period of our sample. Default rates for portfolio loans over
this period were lower (2.1 %) than for the loans in CMBS. As we mentioned
above, while this result at first appears to be counter to our hypothesis, it is
important to account for the impact of renegotiation of distressed loans. Table 4
also shows that portfolio loans were far more likely to be extended over this period
(10.1 %) than securitized loans (1.7 %), consistent with Downs and Xu (2014). 14
This difference is even more pronounced when we limit the analysis to defaulted
loans. One-third (33.9 %) of defaulted portfolio loans were extended compared to a
small fraction of defaulted securitized loans (1.9 %).

13
Originations in 2012 are limited to those loans originated in 2012Q1.
14
The definitions of extension and default in Table 4 are independent of each other, they are not mutually
exclusive categories.
From Origination to Renegotiation

90.00%
80.00%
70.00% CMBS
60.00% Bank - Fixed
50.00%
40.00%
30.00%
20.00%
10.00%
0.00%
1 2 3 4 5 6 7 8 9 10 11+
Original Term in Years
Fig. 4 Distribution of CMBS and bank loans by original term. The blue bars show the distribution by the
original term of CMBS loans as of the first quarter of 2012. The red bars show the distribution of the original
term of bank loans on income producing properties over the same period. Both samples are limited to fixed-
rate loans from the same set of banks active in origination for both securitization and their own portfolios.
Source: Federal Reserve (FR Y-14) and Morningstar

Empirical Methodology

Commercial real estate loans are exposed to risk as economic conditions change over
time. The possibility of financial distress has implications for both loan origination and
renegotiation. In the remainder of the paper, we analyze bank loans and securitized
loans with a focus on the likelihood of distress and the ability and incentive of the
lender to renegotiate in distress.
For our analysis it is important to remember that renegotiation can occur prior to
default or after default. Once a loan enters distress, the lender may choose to proac-
tively amend the terms of the loan in an effort to avoid costly default. A loan
renegotiation can be a simple extension, a partial write-down, or any alteration of the
original terms that mitigates the distress. Of course, once a loan has fallen into default,
the lender once again has the option to renegotiate if the distress costs (or liquidation
costs) are perceived to be too high.

45.00%
40.00%
CMBS
35.00%
30.00% Bank - Fixed

25.00%
20.00%
15.00%
10.00%
5.00%
0.00%
Retail Industrial Hotel Multifamily Office
Fig. 5 Distribution of CMBS and bank loans by property type. The blue bars show the distribution of
property type among CMBS loans as of the first quarter of 2012. The red bars show the portfolio share by
property type among banks loans on income producing properties over the same period. Both samples are
limited to fixed-rate loans from the same set of banks active in origination for both securitization and their own
portfolios. Source: Federal Reserve (FR Y-14) and Morningstar
 L. Black et al.

Table 3 Share of CRE securitized by year. The table reports the distribution by year of origination for CMBS
and bank loans that were still active in 2012:Q1. Both samples are limited to fixed-rate loans on income-
producing properties from the same set of banks active in origination for both securitization and their own
portfolios. It is important to note that this is a censored database, and does not include either CMBS or bank
loans that were resolved, either through prepayment or liquidation, prior to 2012:Q1

CMBS Banks CMBS Share

2000 1114 207 82.3 %

2001 300 67 81.7 %
2002 1263 147 89.6 %
2003 1984 225 89.8 %
2004 2192 259 89.4 %
2005 3784 360 91.3 %
2006 4136 531 88.6 %
2007 3927 1057 78.8 %
2008 56 682 7.6 %
2009 13 287 4.3 %
2010 159 532 23.0 %
2011 538 734 42.3 %
2012a 25 293 7.9 %

Source: Federal Reserve (Y-14)

a
Originations in 2012 are limited to those originated in 2012:Q1

Banks differ from arm’s-length investors in their ability to renegotiate loans in

distress. The legal constraints governing lenders’ actions are different. CMBS loans
cannot be modified until after they have been transferred to a special servicer, which
only occurs after or immediately prior to a default. Special servicers are also bound by
both the Real Estate Mortgage Investment Conduit (REMIC) tax rules and the pooling
and servicing agreements (PSAs) for each specific deal. In addition the special servicer
must Bmaximize recovery to the certificate holders as a collective whole on a net

Table 4 Defaults and extensions of CMBS and bank loans. The table below compares the propensity of loans
that were current as of 2012:Q1 to default over the next ten quarters (until 2014:Q3) for CMBS and for bank
loans. Both samples are limited to fixed-rate loans on income-producing properties from the same set of banks
active in origination for both securitization and their own portfolios. We also report the share of loans extended
for both loans that defaulted over this period and loans that remained current over this period

CMBS Banks

Percent Subsequent Defaulta,c 4.9 % 2.1 %

Percent Subsequent Extendedb,c 1.7 % 10.1 %
Percent Subsequent Extended, Defaulted Loans Only 1.9 % 33.9 %

Source: Federal Reserve (Y-14)

a
Default is defined as the transition of a loan to 60-days past due
b
Extension is defined as increase of more than 3-months in the maturity date of a loan
c
The default and extension definitions are determined independently, they are not mutually exclusive
From Origination to Renegotiation

present value basis,^ (Stafford et al. 2010). The special servicers therefore have to
consider the impact of the interruption of coupon payments to senior tranche holders as
well as potential realized losses to holders of junior tranches. This is a significantly
different standard than the ones facing banks attempting to resolve distressed loans.
Alternatively, banks may be better able to assess liquidation values, or local eco-
nomic conditions, managerial talent, or any type of information that would lead to
better decisions in case of distress.15 In addition, the incentives may differ. Even when
special servicers have the legal authority to renegotiate, it may not be in their interests.
These differences in lenders provide empirical predictions about the funding structure
of projects (An et al. 2011) and the actions of lenders in the event of distress. We
examine the following two predictions.
Our first prediction is that banks fund riskier loans on balance sheet. Controlling for
observable risk characteristics, ex-ante loan rates should be higher for these lenders
with comparative advantage at producing information compared to the less
information-sensitive lenders. Ex-post, financial distress will be higher for this group
as well.
Our second prediction is that banks are more likely to extend loans in distress. Banks
may proactively avoid default by extending loans prior to default. In addition, follow-
ing default, banks should be more likely to extend loans rather than liquidate. This
could either reflect a greater ability of banks to renegotiate or a legal constraint among
servicers of CMBS deals.

Regression Specifications

The empirical analysis begins with an examination of the probability of a loan being
securitized, conditional on the loan being active (including loans in various stages of
default) as of 2012:Q1. Before we can consider the effects of securitization, we need to
understand the characteristics of loans associated with securitization. To do so, we will
use the following specification for the probability of loan i that is currently active as of
2012:Q1 being securitized by bank j:


P Loan Securitized ij ¼ f ðlog of loan amounti ; LTV at originationi ; debt yield at originationi ;
spread i ; property type fixed effectsi ; bank fixed effects j ;
origination year fixed effectsÞ
ð1Þ

where P(•) indicates probability and i and j indicate loan and bank respectively. The
dependent variable is a dummy variable that is one if the loan was securitized in our
sample horizon. This specification uses origination characteristics when available as
well as a few other conditions that reflect the general risk characteristics of the loan;
however, the analysis is cross-sectional at origination. We also include the spread of the
interest rate on the loan to the rate on a treasury security with the same maturity at the
time of origination. The bank fixed-effects control for any differences in propensity to
securitize across banks and the origination year fixed effects control for aggregate
changes in market conditions across quarters.

15
This relates to differences between decentralized and hierarchical firms (Stein 2002).
 L. Black et al.

The second part of the empirical analysis is an examination of the probability of a

loan defaulting. Given that there are observable differences between CRE loans that are
retained and those that are sold, the default likelihood will also differ across these loans.
In addition, there may be unobservable differences across portfolio and securitized
CRE loans that could increase or decrease likelihood of default. For this analysis, we
use the following specification:


P Loan Defaultsij ¼ f S ij ; log of current balancei ; property type fixed effectsi ; current LTV i ;
current occupancy ratei ; debt yield i ; spread

i ; bank fixed
effects j ; origination year fixed effectsi
ð2Þ

where Sij is dummy variable indicating whether the loan was securitized or not. We
estimate this model over a sample of the loans that were current in their payments as of
2012:Q1. The right hand side variables is an indicator for whether the loan defaulted
(i.e. transitioned to a 60 days past due status) at any time from 2012:Q1 to 2014:Q1.
We include an indicator variable for securitization to test whether securitized loans are
more or less likely to default than retained loans. The hypothesis of market segmentation is
that banks specialize in retaining risky loans that do not have the standard features of more
vanilla financial instruments. Under this hypothesis, the coefficient on Sij is predicted to be
negative, because securitized loans should be less likely to default. The alternative is the
adverse selection hypothesis. Banks’ advantage in information production may imply that
banks use private information to Bcherry pick^ good loans and sell bad ones. Under this
hypothesis, the coefficient on Sij is predicted to be positive, because securitized loans are
likely to suffer a lemons problem and should be more likely to default.
To account for renegotiation prior to default (as shown in Table 4), we estimate an
alternate specification of our default model where the dependent variables is whether
the loan defaulted or was extended. We also explore the different degree to which the
original interest rate captured the credit risk between bank and CMBS loans by
including a specification that interacts the spread variable with the securitization
dummy. If bank loans either were riskier, or had pricing that was more sensitive to
credit risk, than CMBS loans this coefficient should be negative.
The analyses on probability of securitization and default will lead us to our primary
analysis on loan renegotiation. The aftermath of the financial crisis resulted in a
significant number of defaults in commercial real estate. We will focus on the period
beginning in the first quarter of 2012 to examine loan renegotiation in the event of loan
distress using a sample of loans that defaulted post 2012, resulting in a sample of 114
portfolio loan and 946 securitized loans.
For our third specification, we look at the probability of renegotiation conditional on
default. This specification takes the following form:

PðLoan Renegotiaioni jDefaultsi Þ ¼ f D^ ij ; S ij; log of current balancei ; property type fixed
effectsi ; LTV at originationi ; change in LTV since
originationi ; current occupancy ratei ; debt yield at
originationi ; change in debt yield since origination
i ; spread i ;
bank fixed effects j ; origination year fixed effects
ð3Þ
From Origination to Renegotiation

where Sij is the securitization indicator variable. In some specifications we will also
include the variable D ^ ij which is the predicted probability of a loan going into default
from Eq. (2). We include this term as an alternate specification to explore some of the
sorting incentives that we spoke to at the onset. Risky borrowers may be naturally
attracted to bank finance because of the implicit option to renegotiate in case of default.
In this case, default is Bexpected^ by the lender and we would expect a positive
coefficient on D^ ij.
Using this analysis, we expect to find that commercial banks are more likely to
renegotiate loans in default. This provides borrowers with a unique value, because the
flexibility of the bank may prevent immediate liquidation upon default. The analysis
may also shed light on important issues discussed during the crisis, such as the role of
securitization in facilitating the problems in the CRE market.

Results

In this section we provide estimates of the models expressed in Eqs. (1)–(3). All models
are estimated with logit specifications. For the securitization model we use the sample
of CRE loans originated by the set of large banks that submit data to the FR Y-14 data
set. While we do find differences in loan characteristics between securitized and
portfolio loans, it is helpful to remember that we have already filtered out some of
the differences between the two loan types when we narrowed the sample to fixed-rate
income-producing loans.
The dependent variable is a binary variable taking value of one if the loan is securitized
(and, thus, observed in the Morningstar data), and zero otherwise. The originations data
span the 2000–2012 period. Although, as mentioned above, we do have a left-censoring
problem in that we do not observe portfolio loans that may have been originated in the
early 2000s but paid off before our sample collection period began.
The explanatory variables include loan-specific risk factors such as the loan size, the
debt yield at origination, the LTV at origination and the type of property backing the
loan. We also include the spread of the loan rate over the comparable Treasury yield
that prevailed at the time of origination. This latter variable helps to proxy for risk
factors that are unobserved by the econometrician, but are observable to the lender and
are priced. All our specifications contain a full set of year dummies and lender fixed
effects.
The results are in Table 5. We have more than 22,000 observations on securitization
choices. Unconditionally, nearly 80 % of the sample was securitized. All specifications
are estimated without a constant. The dispersion of the estimated coefficients on the
property type indicators show that the property type is an important determinant of the
ultimate source of CRE funding. Holding all variables in the model constant, loans on
retail properties (e.g., malls and shopping complexes) and hotels are more likely to be
securitized. Loans backed by industrial, multi-family, and office are relatively more
likely to be retained in the banks’ portfolios.
LTVat origination is positively associated with securitization. This result goes against
our prior that banks would specialize in holding riskier loans, which would tend to have
higher values for these variables. It is possible that this reflects endogeneity in the loan
characteristics data. For a fundamentally risky loan where future cash flows or market
 L. Black et al.

Table 5 Probability of securitization. Logistic regressions of the probability of securitization conditional on

loans being active in 2012:Q1

(i) (ii) + bank (iii) + market (iv) 10-yr. (v) - spread

fixed effects drivers mortgages only

Intercept −17.893*** −17.530*** −18.522*** −17.896*** −19.594***

(.500) (.540) (.852) (1.376) (.830)
Balance Outstanding (log) 1.061*** 1.055*** 1.063*** 1.038*** 1.094***
(.028) (.030) (.032) (.052) (.031)
Debt Yield at Origination 1.070** 1.144** 1.027** 2.805*** .974*
(.440) (.468) (.0510) (1.068) (.516)
LTV at Origination 2.270*** 2.636*** 2.429*** 4.575*** 2.418***
(.183) (.192) (.202) (.371) (.201)
Rate Spread −31.047*** −23.040*** −19.202*** −102.8***
(3.197) (3.269) (3.398) (7.857)
Retail .771*** .807*** .753*** .812*** .759***
(.069) (.072) (.074) (.103) (.074)
Industrial −.071 −.174* −.203** −.328*** −.211**
(.127) (.090) (.093) (.125) (.092)
Hotel .706*** .860*** .821*** .977*** .778***
(.127) (.137) (.147) (.221) (.147)
Multifamily −.272*** −.289*** −.387*** .577*** −.359***
(.071) (.073) (.076) (.129) (.075)
Unemployment Rate −.072*** −.052*** −.072***
(.011) (.018) (.011)
Year-over-Year House 1.512*** 3.280*** 1.472**
Price Appreciation (.610) (.953) (.608)
Bank Fixed Effects No Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes
Number of Observations 22,424 22,424 20,607 15,505 20,607

Standard errors are reported in ()s. * Significant at 10 % level. ** Significant at 5 % level. *** Significant at
1 % level. Office is the omitted property type

conditions are uncertain, lenders may require higher borrower equity, for example, in
order to make and hold a loan in portfolio. The negative coefficient on the loan rate
appears to confirm this intuition. Conditional on the other variables in the logit model,
rates on loans retained by the banks tend to be higher than rates on securitized loans. As
we mentioned before this result may reflect differences in the cost of funds between the
CMBS and bank market, or the presence of either greater risk or more sensitive risk
based pricing in the bank market. The results persist when we limit our sample to just
loans with 10-year terms, in the fourth column.16 The results also persist when we drop
the spread variable, in the fifth column.17
While the securitization model gives some indication that portfolio loans are riskier
than securitized loans in an ex-ante sense, the best test of this is to look at actual defaults.
16
This robustness test was made in response to a comment from a discussant, as almost all CMBS loans in our
analysis are 10-year term loans.
17
An editor suggested this specification in case the pricing of the loans was endogenous to the securitization
decision,
From Origination to Renegotiation

In Table 6 we estimate the default model outlined in Eq. (2). Note again that we see a
strong positive association between default and the interest rate spread on the loan.
Not too surprisingly, traditional risk factors such as the updated LTV and the
occupancy rate have the expected signs. Properties with comparatively higher occu-
pancy rates and lower LTVs had lower default rates. At least in columns (i) and (ii) we
see a positive correlation between the local unemployment rate and loan default.
In columns (i) and (iii) the coefficient on the securitization dummy variable is
positive and statistically significant. This result fits with the data in Table 2 showing
that portfolio loans had lower unconditional default rates and holds when we drop the
spread in column (iii). However, this result is at odds with our finding that loan spreads
were negatively related to securitization, controlling for observable risk factors.
One possibility for this result is that bank loans really are riskier, as the pricing data
would suggest, but retained bank loans have lower conditional default rates because
default is avoided by a loan extension or a renegotiation. To address this possibility we
augment the definition of a default to include both recorded defaults and (for portfolio
loans) current loans that we identify as having been extended. These results are in
columns (iv) and (vi) of Table 6. The coefficient on the securitization variable switches
from positive to negative with this measure of loan distress in column (iv) and when we
drop the spread in column (iv). The securitization dummy in column (iv) has a coefficient
of −0.305, which translates to an odds ratio of 0.737 indicated that securitized loans were
substantially less likely to become distressed. While the securitization dummy in column
(v) is insignificant, the difference in the sensitivity of initial interest rate and subsequent
distress is even more pronounced. This correlation between default and the risk premium
in the original loan pricing appears to be confined to portfolio loans, as evidenced by the
negative sign on the interaction term of the rate spread x securitization. This suggests that
even if portfolio loans are ceteris paribus less likely to become distressed, portfolio loans
where the lender did require a higher initial contract rate are more likely to become
distressed than similarly priced securitized loans. The result may either suggest a higher
sensitivity to the pricing of risk in retained loans or a wider range in retained loans of
underlying unobserved risk characteristics that are not reflected in the reported under-
writing measures (but are reflected in the pricing). It also suggests that the negative sign
for the spread variable in the securitization results may be driven by this difference in risk
or risk sensitivity and not differences in the costs of funds.18
We draw two conclusions from these developments. First, since it is unlikely that a
bank would extend or renegotiate a loan that is not impaired in some way, we interpret
the result as confirmation that the incidence of financial distress in the retained loan
portfolio is higher than in the securitized portfolio. Second, this result speaks to an
important role that banks play as financial intermediaries. Default is costly for both
borrowers and lenders. Banks appear to be able to achieve lower default rates than in
the securitized portfolio not because of better underwriting or screening, but by
avoiding the default state altogether when financial distress occurs.19

18
As a robustness test we re-estimate all these equations replace the securitization dummy with the fitted
probability of securitization from column (iii) on Table 5. All the results persist in this specification.
19
To test this hypothesis that banks engage in efficient recontracting for distressed loans in their portfolio we
would ideally compare ultimate loss rates for retained and securitized loans. Unfortunately we do not observe
losses or recoveries in the bank data over a sufficient time period.
Table 6 Probability of default or distress. Logistic regressions of the probability of default conditional on loans being current in 2012:Q1 and never previously extended. The current
values are as of 2012:Q1

(i) Default (ii) Default + CMBS*spread (iii) Default - spread (iv) Default/Extend (v) Default/ (vi) Default/
Extend + CMBS*spread Extend - spread

Intercept −6.058*** −6.290*** −5.013*** −4.486*** −4.832*** −3.142***

(.500) (.943) (1.145) (.916) (.926) (.886)
Loan Securitized .751*** .943*** .733*** −.305*** .218 −.334***
(.186) (.334) (.187) (.106) (.188) (106)
Balance Outstanding (log) .065* .063 .045 .031 .023 −.001
(.038) (.039) (.038) (.031) (.031) (.031)
Current Debt Yield −7.059*** −7.051*** −7.008*** .168 .132 .139
(.906) (.905) (.902) (.435) (.435) (.431)
Current LTV 2.888*** 2.892*** 2.868*** 2.078*** 2.108*** 2.066***
(.383) (.383) (.384) (.255) (.256) (.255)
Current Occupancy Rate −1.669*** −1.670*** −1.669*** −1.291*** −1.300*** −1.277***
(.154) (.154) (.154) (.130) (.131) (.130)
Rate Spread 21.113*** 28.159** 26.469*** 39.697***
(7.120) (12.370) (4.475) (5.856)
Rate Spread * Loan Securitized −9.905 −28.489***
(14.101) (8.324)
Retail −.756*** −.757*** −.765*** −.597*** −.606*** −.605***
(.089) (.089) (.089) (.075) (.075) (.075)
Industrial −.525*** −.525*** −.527*** −.313*** −.313** −.314***
(.147) (.147) (.147) (.111) (.112) (.111)
Hotel −.642*** −.638*** −.602*** −.579*** −.551*** −.517***
(.163) (.163) (.162) (.134) (.134) (.133)
Multifamily −1.072*** −1.075*** −1.012*** −.735*** −.755*** −.761***
(.113) (.113) (.113) (.088) (.088) (.088)
Unemployment Rate .054*** .054*** .053*** .005 .005 .006
(.016) (.016) (.016) (.013) (.013) (.013)
L. Black et al.
Table 6 (continued)

(i) Default (ii) Default + CMBS*spread (iii) Default - spread (iv) Default/Extend (v) Default/ (vi) Default/
Extend + CMBS*spread Extend - spread

Year-over-Year House Price Appreciation .559 .564 .573 1.181* 1.213* 1.227*
(.822) (.822) (.822) (.674) (.675) (.672)
Bank Fixed Effects Yes Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes Yes
Number of Observations 15,388 15,388 15,388 15,388 15,388 15,388
From Origination to Renegotiation

Standard errors are reported in ()s. * Significant at 10 % level. ** Significant at 5 % level. *** Significant at 1 % level. Office is the omitted property type
 L. Black et al.

To explore the loan extension decision more carefully we next estimate the loan
extension model in Eq. 3. The results are in Table 7. The model is estimated on the set
of loans that were current as of 2012:Q1 and entered default over the next ten quarters
(hence the much smaller sample size). We estimate two different specifications. The
results in columns (i) to (iii) of Table 7 are based on logit models. With the fairly small
number of observations we were unable to get the logit models to converge when we
included bank and time fixed-effects. Thus, for robustness, we also estimated linear
probability models with the complete set of fixed-effects in columns (iv) and (vi).
Again, we also include specifications dropping the spread variable as a robustness test.
We were not able to find observable variables that consistently account for the loan
extension event. Indeed, the only reliable predictor of extension is whether the loan was
retained in the bank portfolio. To be sure, this last result could be due to the fact that
CMBS servicers face different contractual obligations to their investors and are simply
not allowed the same flexibility to renegotiate as banks.
Interestingly, in column (vi) we see that the predicted probability of default from
Eq. 2 has a strong positive coefficient in the extension regression. Thus, even for this
sample where every loan actually defaulted, loans that were ex ante viewed as more
likely to default turn out to have a higher rate of loan extension and renegotiation. This
could be due to the fact that predicted default probabilities in this setting are high
because market conditions have deteriorated (e.g., LTVs have gone up), and lenders
condition on other, borrower and loan-specific terms when making their decision.
However, the relationship between occupancy and extension is not precisely estimated
to confirm this interpretation. It could also be the case that high default probabilities
signal very low recovery values and lenders are simply gambling on keeping the loan
alive and hoping for an improvement in economic conditions. Unfortunately our other
loan-specific variables (debt yield and LTV) are not estimated precisely enough to
disentangle what it is about high default probability loans that makes them more likely
to be extended.
To summarize, the results in Tables 5, 6 and 7 are consistent with our basic
narrative that banks specialize in making risky loans and managing them more
closely in the event of default. Loans that end up in the loan portfolio tend to have
higher rates even though many of the most important risk factors appear to be at
safe levels at the time of origination. The interpretation is that banks demand high
rates in compensation for risk and also take steps to manage these risks right from
the point of origination. The risky loans retained by the banks are more likely to
default. But in the case of default, banks appear to be more willing to work with
borrowers and extend the loan.

Robustness Check with Matched Sample

One potential concern for the analysis is the possibility of endogeneity between the
outcomes of interest (i.e., securitization, default, and extension) and the loan charac-
teristics that serve as control variables in our models (i.e., LTV, debt yield, and
maturity). When a commercial real estate loan is originated, banks may structure the
terms of the loan based on whether the loan is going to be held in portfolio or
securitized. For instance, banks may elect to structure a loan with a fixed interest rate
when there is a high probability of securitizing the loan.
Table 7 Probability of extension. Regressions of the probability of extension conditional on loans that were current in 2012:Q1 and never previously extended and default over the next
ten quarters. The current values are taken at the time of default

(i) Logistic (ii) Logistic + Probability of Default (iii) Logistic – spread (iv) OLS (v) OLS + Probability of Default (vi) OLS - spread

Intercept 9.660 10.824 11.943** 1.018*** .958*** 1.054***

(5.919) (7.166) (6.404) (.219) (.242) (.218)
Loan Securitized −2.533*** −3.234*** −3.372*** −.127*** −.249*** −.253***
(.627) (.786) (.673) (.040) (.055) (.055)
Probability of Default 4.265 4.274 .320** .327**
From Origination to Renegotiation

(3.396) (3.399) (.158) (.157)

Balance Outstanding (log) −.377 −.447 −.474 −.009 −.001 −.003
(.306) (.350) (.341) (.008) (.008) (.008)
Debt Yield at Origination −8.601 3.340 2.404 −.132 .222 .168
(6.748) (7.852) (7.412) (.118) (.409) (.401)
Change in Debt Yield Since Origination .124 .354 .332 .010 .025 .025
(.450) (.494) (.490) (.022) (.023) (.023)
LTV at Origination −2.582 −3.123 −3.234 −.249** −.265*** −.273**
(1.717) (2.010) (1.988) (.089) (.101) (.100)
Change in LTV Since Origination −.917 −3.532** −3.616** .171** −.167* −.168*
(1.082) (1.528) (1.507) (.074) (.095) (.095)
Current Occupancy Rate −.765 −.353 −.503 −.064* −.005 −.010
(.788) (1.171) (1.083) (.039) (.040) (.040)
Rate Spread 22.425 7.491 1.078 1.356
(18.164) (22.058) (1.33) (1.313)
Retail −.009 .014 −.014
(.016) (.018) (.018)
Industrial .014 .055* −.056*
(.029) (.029) (.029)
Hotel −.056* −.041 −.036
(.033) (.034) (.033)
Table 7 (continued)

(i) Logistic (ii) Logistic + Probability of Default (iii) Logistic – spread (iv) OLS (v) OLS + Probability of Default (vi) OLS - spread

Multifamily −.014 −.017 −.001

(.069) (.085) (.083)
Unemployment Rate −.076 −.130 −.132 −.003 −.005* −.005*
(.079) (.098) (.097) (.003) (.003) (.003)
Year-over-Year House Price Appreciation −.726 1.059 1.030 −.094 −.041 −.042
(2.682) (3.015) (3.005) (.102) (.096) (.096)
Bank Fixed Effects No No No Yes Yes Yes
Year Fixed Effects No No No Yes Yes Yes
Number of Observations 638 588 639 638 588 589

Standard errors are reported in ()s. * Significant at 10 % level. ** Significant at 5 % level. *** Significant at 1 % level. Office is the omitted property type
L. Black et al.
From Origination to Renegotiation

In this section, we use a propensity score matching technique to mitigate this problem.
Propensity score-matching methods correct for sample selection bias due to observable
differences between treatment and control groups (Dehejia and Wahba 2002). For our
analysis, we compute propensity scores in order to match retained portfolio loans with
similar securitized loans. The matched sample excludes all observations from the data set that
were not matched. In this way, propensity score matching produces a more direct comparison
between portfolio and securitized loans for analyzing default and extension behavior.
The first stage of matching uses the logit specification for the probability of
securitization in Eq. (1).20 The propensity scores are the fitted probabilities from this
logit model. We then select one-to-one matches from the two groups of loans –
portfolio and securitized – based on a nearest neighbor approach. This approach results
in a sample with fewer observations but greater similarity in observable characteristics.
As can be seen in Fig. 6a, the distribution of fitted probabilities of securitization differs
substantially for loans that were held in portfolio (blue dashed line) compared to loans
that were securitized (red solid line). In contrast, Fig. 6b shows that the propensity score
matching process generates a subsample of portfolio and securitized loans with a
similar distribution of securitization probabilities.
With this matched sample, we re-estimate the default model in Eq. (2). The results of
this exercise are in Table 8. As we can see, the sample size drops markedly from over
16,000 observations in the unmatched sample in Table 6 to just over 2000 observations.
The results generally confirm our story about borrower risk and bank behavior. In
columns (i) through (iii) of Table 8, where the definition of default is simply whether or
not the loan is 90-days past-due, we now get no significant differences in default rates
between portfolio and securitized loans. In columns (iv) through (vi) of Table 8 we
employ the expanded definition of distress that includes defaulted loans and loans that
were extended prior to maturity. The specifications in columns (iv) and (v) bear out our
earlier result that distress is more likely for retained loans than securitized loans. The
coefficient of −0.4136 on the securitization dummy (column v) translates to an odds ratio
of 0.66, indicating that securitization is associated with a lower rate of distress that is both
statistically and economically significant. While the coefficient on securitization alone in
column (vi) is positive (but insignificant), when the rate spread is evaluated at the mean
level (.018) we see that the combined or total effect of securitization in column (vi) is
strongly negative. Therefore, the propensity score matching exercise confirms our previ-
ous findings that banks retain riskier commercial real estate loans. This indicates that our
baseline results are robust to concerns about endogeneity in the loan origination process.
Due to sample size, it is not possible to apply our propensity score matching
approach to the renegotiation model in Eq. (3), where we look at the loan extension
probability conditional on default. The sample size is already quite small at about 600
observations in Table 7.

Conclusion

Securitization of commercial real estate loans has grown dramatically over the past
20 years. Despite this growth, a substantial majority of the commercial mortgages

20
Specifically, we use the specification that generates the results in column (iii) of Table 5.
 L. Black et al.

a
Distribution of Propensity Scores, Full Sample
15
10
5
0

0 .2 .4 .6 .8 1

Not securitized Securitized

b
Distribution of Propensity Scores: Matched Sample
15
10
5
0

.2 .3 .4 .5 .6 .7

Not Securitized Securitized

Fig. 6 Distribution of propensity scores. Figure 6a shows the distribution of propensity scores for the full
sample and Fig. 6b shows the distribution of propensity scores for the matched sample. The dashed blue line
indicates loans not securitized (portfolio) and the red line indicates securitized

outstanding still remains on the balance sheets of banking institutions. This outcome
reflects an equilibrium where certain types of investors and lenders have comparative
advantage in funding certain types of loans.
We use confidential supervisory loan level data on portfolio loans collected as part
of the recent bank stress tests in the U.S. In conjunction with loan-level data from
commercial mortgage-backed securities, we can explore the differences between port-
folio and securitized loans.
Table 8 Default models with matched sample. Logistic regressions of the probability of default. All observations conditional on loan being current in 2012.Q1. Sample is based on
propensity score matching propensity score matching

(i) (ii) (iii) (iv) (v) (vi)

Default Distress
Default + Extension

Constant −5.0068 −6.7931* −6.7725* −3.1903 −4.7971* −4.8622*

(3.6427) (3.6581) (3.6592) (2.8163) (2.8832) (2.9215)
From Origination to Renegotiation

Securitized 0.1786 0.2385 0.4334 −0.4333** −0.4136** 0.2866

(0.2784) (0.2766) (0.5087) (0.1866) (0.1814) (0.3762)
Balance outstanding (log) −0.0099 0.0601 0.0529 −0.1195 −0.0289 −0.0430
(0.1578) (0.1605) (0.1599) (0.1196) (0.1239) (0.1245)
Debt yield at origination −13.0255*** −13.6592*** −13.6494*** −2.6035 −2.8719* −2.9650*
(3.2232) (3.2847) (3.2779) (1.6291) (1.6852) (1.6794)
Current LTV 1.2230 1.1201 1.1050 1.7174*** 1.5634*** 1.5574***
(0.8343) (0.8141) (0.8219) (0.5811) (0.5749) (0.5807)
Spread 42.3805*** 45.1132*** 35.9490*** 45.0821***
(13.2136) (14.4045) (8.9809) (11.0531)
Spread × Securitization −10.8346 −37.3623**
(23.3800) (16.8923)
Current Occupancy Rate −1.1539*** −1.1430** −1.1345** −0.7240** −0.7148** −0.7478**
(0.4447) (0.4491) (0.4479) (0.3477) (0.3486) (0.3430)
Unemployment Rate 0.0801** 0.0829** 0.0836** −0.0159 −0.0169 −0.0130
(0.0404) (0.0410) (0.0409) (0.0361) (0.0371) (0.0369)
4Q Chg in HPI 5.3808** 5.7041** 5.7390** 3.7769* 3.7670* 3.9256*
(2.6335) (2.6682) (2.6784) (1.9597) (2.0113) (2.0386)
Table 8 (continued)

(i) (ii) (iii) (iv) (v) (vi)

Default Distress
Default + Extension

Bank fixed-effects yes yes yes yes yes yes

Property type controls yes yes yes yes yes yes
Year dummies yes yes yes yes yes yes
N 2081 2081 2081 2201 2201 2201
pseudo R-sq 0.1511 0.1675 0.1677 0.1077 0.1229 0.1266

Robust standard errors in parentheses. * Significant at the 10 % level. ** Significant at the 5 % level. *** Significant at the 1 % level. Office is the omitted property type
L. Black et al.
From Origination to Renegotiation

This newly available bank loan data highlights the substantial differences between
portfolio and securitized commercial real estate loans. Loans in CMBS deals are almost
entirely fixed-rate loans on stabilized income-producing properties, whereas bank loans
include floating-rate loans, construction loans, and other types. This suggests that banks
may have a comparative advantage in funding Bnon-standard^ loans.
The analysis in this paper focuses on loan distress and renegotiation. Banks appear to
have a comparative advantage in renegotiation that yields two predictions: banks fund
riskier loans and banks are more likely to extend loans in distress. Based on an overlapping
sample of similar portfolio and securitized loans, we develop our analysis in three steps: the
likelihood of securitization, the likelihood of distress, and the likelihood of extension.
We find some evidence that bank lenders specialize in funding riskier CRE loans as
compared to the capital market investors. While ex-ante loan characteristics are not
significantly different across the bank loans and loans in CMBS, this finding likely
reflects the endogenous response by banks to tighten underwriting standards for loans
on properties that are fundamentally riskier than those in CMBS.
We also find that banks are far more active in using loan extensions to mitigate
default risk. The probability of loan extension or renegotiation prior to default and
conditional on default both appear to be higher for bank loans.
These results are consistent with a market segmentation story where borrowers with
risky, difficult to assess projects are matched with lenders with comparative advantage
in renegotiation. The observable differences between portfolio and securitized loans
point to a fundamental difference between these two sources of funding. In particular,
banks appear to have a unique role in managing risk for loans in distress.

Acknowledgments We thank seminar participants at the UF/FSU Real Estate Symposium, the Stress Test
Modeling Research Conference, the Interagency Risk Quantification Forum, the AREUEA Annual Meetings, the
Southern Finance Association Annual Meetings, the Federal Reserve Bank of Cleveland, Ohio University,
University of Cincinnati, as well as Travis Davidson, Ronel Elul, Emre Ergungor, Mark Lueck, Wayne Passmore,
and Gokhan Torna for helpful comments, and Joseph Cox and Erin McCarthy for research assistance.

Appendix Construction of Data Set for Estimation

We draw our loan sample from two distinct data bases, the Morningstar CMBS
database and the FR Y-14 Q CRE schedule. This appendix provides some additional
detail on the construction of the database, including some characteristics of the portions
of both databases that were excluded from the analysis.
The CMBS data is drawn from the Morningstar database, which covers the CMBS
market at the deal level. These deals include loans originated by banks, insurance
companies, conduits and others. As of 2012:Q1, approximately 59 % of active CMBS
loans were originated by banks. We limit our analysis to fixed rate loans originated by
firms that are also in the FR Y-14 collection, which accounts for 84 % of CMBS bank
originated loans. The FR Y-14 data collection included 19 firms in 2012:Q1.21 Within
this sample we restrict our analysis to fixed rate loans originated and held by the 8 firms
that are also active in the CMBS market. The decision to restrict the sample to both

21
Coverage has currently expanded to 32 firms, but we limit our analysis to those firms participating in the
initial reporting wave in order to maximize the time available to monitor loan performance.
 L. Black et al.

fixed rate loans and to the same group of firms active in both the markets was made to
minimize concerns regarding the endogenity of the securitization decision. We limit our
analysis to a group of loans that have similar loan structures and are all from firms that
had the option to either securitize or hold loans on their portfolio.
Appendix Table 9 below compares the portfolio characteristics of our estimation
sample with the portions of the CMBS and bank portfolio that we excluded from our
analysis. The first two columns report the characteristics from banks in the CMBS
sample that are not FR Y-14 reporters and from FR Y-14 reporters that are not active in
the CMBS market. The third and fourth columns report the characteristics of the
adjustable and mixed-rate loans from the sample of nine banks active in CMBS market
that also were FR Y-14 reporters. This table reflects the active portfolios as of 2012:Q1.
The outlier in these six portfolios is clearly the adjustable rate CRE loans in CMBS
portfolio that were originated by our sample of nine banks. As of 2012:Q1 there were
very few of these adjustable rate CMBS loans still active. They were dominated by a
small number of very larger hotel loans, many of which were in default. The negative
spread to Treasuries reflects that many of these loans may have been indexed to
baseline rates other than Treasuries at origination.
Restricting ourselves to the other portfolios, the patterns we observe in our sample
repeat. CMBS loans are larger with higher LTVs ratios and occupancy rates in each set of
samples. Bank loans have higher debt yields and spreads to Treasuries than CMBS loans.
Finally, as of 2012:Q1 more CMBS loans were in default than the bank loan portfolios.

Table 9 Characteristics of excluded and included loans

CMBS – Out Banks – Out CMBS – Banks – ARM CMBS – In Banks – in

of Sample of Sample ARM in in Sample Sample Sample
Sample

Total Loans 25,298 11,944 124 19,742 19,496 5382

Current Balance $ 11.4 $ 6.6 $167.4 $4.9 $13.8 $4.0
($ millions) (28.9) (19.3) (693.8) (11.7) (35.4) (10.1)
Balance at Origination $ 12.8 $ 7.3 $221.4 $5.9 $15.0 $4.4
($ millions) (34.8) (20.4) (787.7) (15.8) (40.7) (10.1)
Current LTV 61.6 % 58.1 % 49.3 % 57.3 % 62.5 % 55.8 %
(14.1) (22.4) (17.8) (17.8) (12.8) (19.7)
LTV at Origination 68.9 % 62.0 % 57.5 % 60.2 % 69.1 % 60.1 %
(13.0) (21.3) (16.3) (18.3) (11.4) (20.5)
Current Occupancy Rate 87.7 % 84.6 % 72.1 % 83.0 % 88.9 % 83.7 %
(19.7) (27.6) (20.2) (25.4) (17.6) (28.5)
Current Debt Yield 13.0 % 17.3 % 16.3 % 13.4 % 13.0 % 15.6 %
(9.3) (15.9) (17.4) (10.9) (9.7) (12.1)
Debt Yield at Origination 12.7 % 16.0 % 22.4 % 12.2 % 11.5 % 13.6 %
(9.7) (15.6) (21.8) (10.2) (5.8) (10.9)
Spread to Treasuries 1.5 % 1.6 % −2.1 % 0.9 % 1.5 % 2.4 %
(0.8) (0.2) (2.4) (2.0) (0.7) (1.5)
Percent in Past Due 15.7 % 6.5 % 82.3 % 4.6 % 14.7 % 5.2 %
Or Previously
Extended
From Origination to Renegotiation

Appendix Fig. 7 below provides additional evidence that the adjustable rate portfo-
lios for the CMBS portfolios are fundamentally different. CMBS adjustable rate loans
from our sample of nine banks have some of the shortest portfolio distributions in our
sample while bank adjustable rate loans have some of the longest. The differences in
the distribution of the original term between bank and CMBS loans between the fixed
rate loans in our sample and the total portfolio from originators outside our sample are
however quite similar.

90.00%

80.00%
CMBS - In Sample
70.00% Bank - In Sample

60.00% CMBS - ARM In Sample

Bank - ARM In Sample
50.00%
CMBS - Out of Sample
40.00% Bank - Out of Sample

30.00%

20.00%

10.00%

0.00%
1 2 3 4 5 6 7 8 9 10 11
Original Term in Years
Fig. 7 Distribution of bank and CMBS portfolios by original term

Appendix Fig. 8 below shows the distribution across our six portfolios by
property type. Again the adjustable rate CMBS loans from our sample of nine
banks represents a strong outlier, being dominated by hotel loans while adjustable
rate loans in bank portfolios are dominated by multifamily loans. The property
distribution for CMBS and bank loans from our sample of nine banks is similar to
the distribution we observe from the originators we exclude from our sample. These
figures support our contention that our sample does provide a representative sample
to examine CMBS vs. bank portfolios, and supports our decision to exclude the
adjustable rate mortgages.
The larger size for CMBS loans may reflect differences in tenant quality, and
therefore credit risk, between CMBS and bank loans. In the retail space the larger
properties, such are often Banchored^ by high quality tenants than those served
by smaller Bstrip^ centers. Unfortunately we lack the detailed information on
tenant quality in both databases. The size of the property, and the related loan
balance, may be used to proxy for the presence of such Banchored^ retail
properties, and similar properties with high quality tenants in other property
types. Appendix Table 10 below reports the differences in the average balance
at origination between CMBS and banks by property type. We see here that
CMBS loans consistently have larger balances at origination than bank loans
across all property types.
 L. Black et al.

70.00%
CMBS - In Sample
60.00% Bank - In Sample
CMBS - ARM In Sample
50.00%
Bank - ARM In Sample
40.00% CMBS - Out of Sample
Bank - Out of Sample
30.00%

20.00%

10.00%

0.00%
Retail Industrial Hotel Multifamily Office
Fig. 8 Distribution of bank and CMBS portfolios by property type

Insert text on how ARM portfolios are different, in particular in CMBS, but out of
sample portfolios are similar.

Table 10 Balance at origination by property type ($ millions)

CMBS Banks

Retail $ 12.9 (33.7) $ 4.4 (10.1)

Industrial $ 9.3 (21.5) $ 3.4 (4.5)
Hotel $ 21.0 (65.6) $ 6.5 (10.1)
Multifamily $ 10.5 (31.8) $ 3.6 (5.2)
Office $ 24.9 (54.1) $ 5.9 (15.7)

References

Agarwal, S., Amromin, G., Ben-David, I., Chomsisengphet, S., & Evanoff, D. (2011). The role of securiti-
zation in mortgage renegotiation. Journal of Financial Economics, 102, 559–578.
Akerlof, G. A. (1970). The market for Blemons^: quality uncertainty and the market mechanism. Quarterly
Journal of Economics, 84, 488–500.
Ambrose, B., & Sanders, A. B. (2003). Commercial mortgage (CMBS) default and prepayment analysis.
Journal of Real Estate Finance and Economics, 26, 179–196.
An, X., Deng, Y., & Gabriel, S. (2011). Asymmetric information, adverse selection, and the pricing of CMBS.
Journal of Financial Economics, 100, 304–325.
Archer, W. R., Elmer, P. J., Harrison, D. M., & Ling, D. C. (2002). Determinants of multifamily mortgage
default. Real Estate Economics, 30, 445–473.
Black, L., Chu, C. S., Cohen, A., & Nichols, J. B. (2012). Differences across originators in CMBS loan
underwriting. Journal of Financial Services Research, 42, 115–134.
Ciochetti, B., Deng, Y., Lee, G., Shilling, J. D., & Yao, R. (2003). A proportional hazard model of commercial
mortgage default with origination bias. Journal of Real Estate Finance and Economics, 27, 5–23.
Coleman, A. D. F., Esho, N., & Sharpe, I. G. (2006). Does bank monitoring influence loan contract terms?
Journal of Financial Services Research, 30, 177–198.
Dehejia, R. H., & Wahba, S. (2002). Propensity score-matching methods for nonexperimental causal studies.
Review of Economics and Statistics, 84(1), 151–161.
From Origination to Renegotiation

Deng, Y., J.M Quigley, and A.B. Sanders (2004). Commercial mortgage terminations: evidence from CMBS.
USC Working paper.
Downs, David H. and Pisun (Tracy) Xu (2014). BCommercial real estate, distress and financial resolution:
portfolio lending versus securitization.^ Working paper.
Esaki, H., S. L’Heureux, and M.P. Snyderman (1999). Commercial mortgage defaults: an update. Real Estate
Finance, 16, 81–86.
Furfine, C. (2010). Deal complexity, loan performance, and the pricing of commercial mortgage backed
securities. Kellogg School of Management working paper.
Ghent, A. and R. Valkanov (2013). Advantages and disadvantages of securitization: evidence from commer-
cial mortgages. Working paper.
Keys, B. J., Mukherjee, T., Seru, A., & Vig, V. (2009). Financial regulation and securitization: evidence from
subprime loans. Journal of Monetary Economics, 56, 700–720.
Piskorski, T., Seru, A., & Vig, V. (2010). Securitization and distressed loan renegotiation: evidence from the
subprime mortgage crisis. Journal of Financial Economics, 97, 369–397.
Puranandam, A. (2011). Originate-to-distribute model and subprime mortgage crisis. Review of Financial
Studies, 24, 1881–1915.
Snyderman, M. P. (1991). Commercial mortgages: default occurrence and estimated yield impact. Journal of
Portfolio Management, 18, 82–87.
Stafford, T., Linder, D., & Jones, R. (2010). CMBS under stress. Frequently asked questions about key
provisions in CMBS pooling and servicing agreements addressing mortgage loan modifications. Real
Estate Finance, 27(1), 3–6.
Stein, J. C. (2002). Information production and capital allocation: decentralized versus hierarchical firms.
Journal of Finance, 57, 1891–1921.
Titman, S., & Tsyplakov, S. (2010). Originator performance, CMBS structures, and the risk of commercial
mortgages. Review of Financial Studies, 23, 3558–3594.
Titman, S., Tompaidis, S., & Tsyplakov, S. (2005). Market imperfections, investment flexibility, and default
spreads. Journal of Finance, 59, 165–205.
Vandell, K. D., Barnes, W., Hartzell, D., Kraft, D., & Wendt, W. (1993). Commercial mortgage defaults:
proportional hazards estimation using individual loan histories. Journal of American Real Estate Urban
Economic Association, 21, 451–480.