1 Title: Bad Loan Externalities: Evidence from the Syndicated Loan Market This version: October 25, 2015 Abstract This study examines external impacts of distressed bank loans on the lending banks and other borrowing firms. The banks, on average, lose almost 1% of their total market value, and the effect spills over to other loan syndicate members. The distress news also impacts the banks’ other borrowers, who experience seven-day mean cumulative abnormal returns of -0.31% for each distress announcement. Distress externalities are worse when the bank is more exposed to the bad loan, and for borrowers that are more relationship dependent. Future lending business is also negatively affected, as loan rates rise by 67 BP following large distress damage, and lenders are less likely to retain existing relationship borrowers. JEL classification: G01; G21; G33. Keywords: Financial intermediation; bank relationships; bankruptcy; lending constraints
54
Embed
Title: Bad Loan Externalities: Evidence from the ... · A related question is the impact on the choice of lead arrangers for new syndicated loans following extensive bad loan damages.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Title: Bad Loan Externalities: Evidence from the Syndicated Loan Market
This version: October 25, 2015
Abstract
This study examines external impacts of distressed bank loans on the lending banks and other
borrowing firms. The banks, on average, lose almost 1% of their total market value, and the effect
spills over to other loan syndicate members. The distress news also impacts the banks’ other
borrowers, who experience seven-day mean cumulative abnormal returns of -0.31% for each
distress announcement. Distress externalities are worse when the bank is more exposed to the bad
loan, and for borrowers that are more relationship dependent. Future lending business is also
negatively affected, as loan rates rise by 67 BP following large distress damage, and lenders are
less likely to retain existing relationship borrowers.
JEL classification: G01; G21; G33.
Keywords: Financial intermediation; bank relationships; bankruptcy; lending constraints
2
1. Introduction
In the Miller-Modigliani (1958) world of perfect capital markets, firms can source capital in
simple transactions, and they are unmoved by shocks at their lending banks. In imperfect
markets, with adverse selection and moral hazard, raising capital and changing its source can
result in complex and costly transactions (e.g., Holmström and Tirole, 1997). Many firms rely on
private bank loans as a source of low cost capital, and many banks supply firms with loans to
earn revenue. Most of the loans are relationship-based as borrowers often return to the same
lender when they need additional capital. Berger and Udell (1998) report the average relationship
lasts 7.8 years, and is longer for small and relatively information opaque firms.
In relationship lending, the bank is at the hub of a web of concurrent relationships with many
borrowers. It provides essential reputation support to the relationship borrowers. Bank reputation
is similar to a Club Good, as it is available only to borrowers in the bank’s relationship circuit.
Reputation is excludable because it is unavailable to borrowers outside the bank’s relationship
web. Within the web, reputation is non-rivalrous as each borrower enjoys the good without
exclusion. By establishing a lending relationship with the bank, each borrower can use the bank’s
reputation support continually, regardless of the number of borrowers in the web. The bank’s
marginal cost of supplying reputation to a new borrower is virtually zero. When the lender
becomes more reputable, each relationship borrower benefits from the improvement. On the
other hand, if the lender’s reputation deteriorates, it can simultaneously impact the relationship
borrowers as well. In addition to reputation support, the bank also provides capital support in
terms of current and future loans. When its resources are reduced due to sudden shocks, it is
unlikely to keep up the same level of capital support, which can in turn hurt the relationship
borrowers. The more relationship-dependent borrowers, such as those with stronger ties with the
3
bank and those that rely heavily on the bank for monitoring, certification, and capital support, are
more likely to be affected by the nexus shocks than are the less dependent borrowers in the
relationship.
This broader recognition of the Club Good nature of the relationship web points to three
important areas that have been neglected in previous empirical studies of lending relationships.
First, earlier studies tend to focus on lending relationships outside the United States. For
example, Gibson (1995) uses Japan data to show lenders’ economic health affects borrowers’
investment behavior, a real external long-term effect beyond immediate losses in equity value.
Second, previous studies tend to focus on macro-level, crises shocks. In response to damages
from Korean bank crises, Bae et al. (2000) report sharp lending pull backs by banks on many
loans. Also, Chava and Purnanandam (2011) show following the 1998 Russian sovereign default,
most affected banks cut back future lending, especially those with more exposure to the
government’s bond default. Santos (2011) finds that after the 2007 to 2009 financial crisis shock,
loan spreads paid by U.S. firms are higher due to a reduction in the supply of total loanable
funds. The latter two studies report the negative impacts on the borrowers, in terms of access to
credit markets and their cost of capital. Evidence from these macro-crises shocks is limited and
should be interpreted with caution. During financial crises, the economic consequences fall
concurrently on all borrowers and lenders in the economy, thus preventing the observation of
spillover effects that could be due to individual loan defaults within the relationship web.1 In
addition, these shocks occur relatively infrequently. For example, Daniel et al. (2013) identifies
only one financial crisis as big as the 2009 crisis over the last century. Third, focusing on
aggregate loan shocks with such broad impacts confounds the analyses of individual bank loans
1 See also Kang and Stulz (2000), Ongena et al. (2000), Brewer et al. (2002), Gan (2003), and Carvalho et al. (2011).
4
because the shocks can affect many aspects of the market at the same time. In that confounded
setting it is challenging to empirically identify and isolate the cause of a loss in firm value or a
preferential loan agreement.
This study, on the other hand, provides a new micro level understanding of the effects of
distressed loan shocks that fall on other borrowers in lenders’ relationship webs and on their
future lending businesses. The purpose is to examine how isolated bad loan shocks, resulting
from individual firm defaults or bankruptcies, simultaneously affect other borrowers in the
relationship bank’s web, and perhaps the structure of the web. The sample of default and
bankruptcy announcements spans multiple decades. The bad loan shocks are relatively isolated
across the economy, so they are unlikely to concurrently move the market. As a result, negative
impacts from individual bad loans shocks that fall on the web of relationship firms will be most
vivid, with few complicating macro level influences.
This paper contributes to the understanding of contagion effects of defaults and bankruptcies
(e.g., Chava and Purnanandam, 2011; Das, Duffie et al. 2007; Murfin, 2012). It extends the
evidence of bad loan externalities to include the consequences of bad loan shocks for all parties
in the syndicated loan market. It reports each party is affected differently by the individual loan
shocks. The degree of impact largely depends on the size of the bad loan, reputation of the
lender, and web borrower’s reliance on the lending relationship. Focusing on the stock
performance of the affected banks and their borrowers alleviates the endogeneity issue that is
common in corporate finance. The default and bankruptcy announcements are often isolated and
localized, so they are unlikely to trigger market-wide catastrophe. When web borrowers
underperform during the announcement periods, it is less likely to be caused by the bad market.
5
To my knowledge, this is the first study that examines the differential impacts of bad loans
on lead arrangers, loan syndicate participants, and their web borrowers. Prior studies do not
examine how syndicate participants in their relationship web are affected by bad loans, and by
how much. In contrast, when a firm defaults on its loan, all lenders in the loan syndicate could be
affected to some degree. By comparing and contrasting the stock returns of lead arrangers and
syndicate participants, the incremental value loss associated with reputation damage can be
isolated from damages from capital depletion (the uncollectable interest and principal on the bad
loan, whose recovery rate is generally very low). The structure of a syndicated loan thus provides
a unique opportunity to disentangle the causes of bad loan externalities, financial constraints vs.
reputation damage, and the value of each. By focusing on isolated individual distress news, the
paper provides new understanding of the economics of lender-borrower connections.
The rest of the paper proceeds as follows. Section 2 outlines the general hypothesis and basic
predictions. Section 3 describes the data and methodology used to construct the sample. Section
4 presents the findings on the impact of loan distress announcement, and Section 5 concludes.
2. Literature review and empirical hypothesis
In the syndicated loan market, lead arrangers often have unique access to information about
their clients through repeat lending and close monitoring (Diamond 1984, Ramakrishnan and Thakor
1984, Fama 1985, Petersen and Rajan 1994). Market participants generally rely on such monitoring
and certification services. Lead banks experience significant value loss when the market realizes
they are unable to properly monitor their borrower (e.g., Dahiya et al., 2003). When an isolated
shock is large enough to impact its financial health and reputation, the web borrowers that are
connected to the lead via lending relationship suffer. For example, Houston et al. (2014) report
the abnormal bond return is negative and significant for the affected borrowers that share the
6
same lead lender as the defaulted borrower. The damage is even stronger for the less-known and
more bank-dependent borrowers. There is no doubt that the lead lender plays an essential role in
the syndicated loan market, but the participants are also important part of it. They are responsible
for providing capital support. The bad loan can make it more difficult for them to lend in the near
future when it is relatively large or when there are multiple bad loans over a short period of time
due to capital depletion. A tightening of the credit market is observed following the financial
crisis (e.g., Bae et al., 2000; Chava and Purnanandam, 2011; Santos, 2011). As a result, their
relationship borrowers will be negatively affected although the effect will be much smaller than
those of the lead arrangers. The characteristics of both borrowers and lenders can have
tremendous influence on the level of impact. For example, when the affected bank is more
reputable, the impact is likely to be smaller because its reputation capital is too strong to be
affected by isolated bad loans. Also, more reputable banks tend to have more capital resources to
absorb the shocks. Thus, large or dominant banks are unlikely to be affected by isolated distress
news (Gopalan et al., 2011). Moreover, for borrowers that have multiple relationship lenders,
they are unlikely to be affected by the downfall of one-single bank because it is easy for them to
switch to different lenders. On the contrary, the bank-dependent borrowers should, in theory,
experience significant negative impact.
Furthermore, there are long-term economic consequences of the bad loans. The direction of
the impact should vary depending on the characteristics of both banks and borrowers. For the
more reputable banks that care more about maintaining existing clientele, they are likely to lower
the loan spread. The more reputable lenders often have higher-quality borrowers based on
positive assortative matching. Following reputation damage, the bank will need to make loan
terms more attractive to keep the high-quality borrowers. By considerably lowering the loan
7
spread on future loans with the web borrowers, the bank is more likely to retain its good clients,
which will in turn help the recovery of its reputation. For the less reputable banks, however, it
may not be the case. Less reputable banks are often associated with the small and more bank-
dependent borrowers. Because the borrowers’ other connections are weak and to establish new
banking relationships is costly, they are locked in with the existing lender. As a result, the
affected bank can extract a higher loan fee from them to cover for the recent financial losses.
This suggests that loan spread will be higher following the distress damage especially for the less
reputable banks and bank-dependent borrowers.
A related question is the impact on the choice of lead arrangers for new syndicated loans
following extensive bad loan damages. Are borrowers more likely to switch to new lenders if
their existing lender experiences significant bad loan damages? Since the borrowers are distinctly
different the answer to the question should vary as well. Borrowers with multiple relationship
banks will incline to choose a different lead arranger for their new loans, as doubt about the
ability of their current lender to provide same quality of monitoring and funding service rises.
The deterioration of the lender’s welfare can no longer satisfy their needs, so the borrowers move
to a new relationship web that is more beneficial to them. However, this is unlikely to be the case
for the smaller and more bank-specific borrowers. Their close relationship with the bank makes
it hard for them to switch to a different lender and start a new lending relationship. Hence, not
only will the probability of switching to a new lender be affected by the degree of bad loan
damages, it also depends largely on borrower’s characteristics and the strength of the existing
lending relationship. Large distress damage is more likely to trigger a change of the lead
arranger, but primarily for the less reputable banks. Firms with strong ties with the current lender
are less likely to establish new lending relationships, ceteris paribus.
8
3. Data and descriptive statistics
The primary sample includes all syndicated loans originated in the U.S. during 1988 through
2012, from Loan Pricing Corporation’s (LPC) DealScan, along with loan related information.
Standard & Poors (S&P) from Compustat rating database is used to identify firm default
announcements. New Generation Research's bankruptcy database is used for information on
Chapter 11 bankruptcy announcements. Center for Research in Security Prices (CRSP) provides
information related to stock price performance. Compustat gives other financial details of the
borrowing firms. The Federal Reserve’s Quarterly Call Report is used for bank related financial
information. Various data sources are then merged together using either established link files or
careful hand matching in the absence of unique and reliable indicator. Details are described
below.
3.1 Loans
I define bad loans as outstanding loans of the firms that appear on either Bankruptcydata.com
or on S&P long term debt rating with a selective default (SD) ad default (D) rating. The
matching between Bankruptcydata.com and DealScan is done manually with firm names, while
the matching between Compustat and DealScan is accomplished with the help of Compustat-
Dealscan link file, which is created by Michael Roberts and WRDS (e.g., Chava and Roberts,
2008). For each bad loans identified following the match, I extract all loan-related contract
information from DealScan, such as the identity of lenders, both lead and participants, loan start
and end date, loan amount, loan type, etc. Each matched loan is categorized as either a default or
a bankrupt loan, which is important for comparison tests. When there are multiple bad loans for
one firm bankruptcy or default news, each loan is separately examined to ensure one loan
observation per bank. Unfortunately, both default and bankruptcy announcements can be
9
partially anticipated. For one, there could be other rating agencies release default information
prior to S&P credit rating, so the market will be informed of the distress prior to my
measurement. For two, there could also be other news reports regarding the underperformances
of the distress firms, so the investors are aware of the high likelihood of distress of certain
borrowers. Under both scenarios, the market adjustment is likely to occur way before my
measurement period, which will make it harder for me to capture any impacts from my chosen
announcements because they are not “news”. As a result, the actual impact of the distress news
should be much larger than what I find due to the partial anticipation.
Following the identification of the lender, I obtain its relationship borrowers as well as their
outstanding loans at the time of the bad loan. Because banks often merge with other banks, its
name and companyid, the bank identifier on DealScan, changes. Manual adjustments account for
these changes. For example, prior to its Merill Lynch acquisition, Bank of America’s companyid
was 84685. After the acquisition, its name becomes Bank of America Merill Lynch, and its new
companyid is 127349. Since both companyids represent Bank of America, when it experiences a
bad loan shock, relationship borrowers are identified using both companyids. When lenders are
bank subsidiaries, the parent company is used for matching and testing. I define the affected
firms as the relationship borrowers of the lender that experiences bad loan shocks. When an
affected firm takes out multiple loans from the same bank, they are aggregated into one
observation.
3.2 Stocks
I use the Compustat-CRSP link to match the companies to CRSP firms in order to get their
stock performances. For those that can’t be identified using the link table, their names are hand-
matched to find more observations. There are quite a few companies and banks that are present
10
on CRSP, so there could be a selection bias, but the bias should go against finding the predicted
results because the banks and firms that have sufficient stock data are usually larger and more
transparent to investors, making them less vulnerable to small shocks. Following Fernando, May,
and Megginson (2012), the main event-window is the seven-day window from 5 days prior to the
distress announcement (both default and bankruptcy) and 2 days after. CARs using other
windows are also reported. All abnormal returns are estimated using Fama-French-Carhart 4-
factor model with information from day-250 to day -50 from the announcement date.
3.3 Variable construction
A key independent variable is Exposure, a continuous measure of the importance of the
distressed firm to the affecting bank. Exposure is constructed by first identifying all outstanding
loans associated with the distressed firm at the time of distress announcement, then aggregating
them at the bank level. Since there are often multiple lenders on the deal, I multiply the DealScan
loan allocation variable by the total facility amount to obtain each lender’s loan proportion.
When the detail information about loan allocation is unavailable, the total facility amount is
divided by the number of lenders.
Lender sizes are quite different. Larger banks have more buffers to absorb the loan shocks, so
it takes a much bigger distress to make an impact. Therefore, the distress loan amount must be
scaled to control for the size effect. To successfully scale the number, the total distress amount is
divided by the average loan amount originated by the bank over the prior two years to get
Exposure, which is in percent. A higher Exposure means the affected bank has more exposure to
the distressed firm, capital-wise. Hence, banks with higher Exposure to the distressed firm are
expected to have more negative stock return reactions to the distress news.
11
The variable that identifies recent distress loans (Surprise_distress) is also important to the
tests. It equals 1 when the distress happens within the first two years of loan origination and 0
otherwise. For the lead arrangers, as time passes it becomes harder to attribute bad loans to their
inadequate screening and monitoring since market conditions are likely to have changed since
the time the loan was originated. A distant default is less informative than a recent one. For the
syndicate participants, banks can sell their loan shares after the loan origination, which can shield
them from the bad loan impact. Even if the lenders decide to hold onto the loan until maturity, as
time elapses, they are able to recover more and more initial investment. Therefore, when the
distress loan is further down the road, its impact on other parties should be less.
A complete list of variable definitions is in Appendix.
3.5 Summary statistics
Table II Panel A provides an annual summary of the new, distressed, and affected loans,
banks, and firms, from January 1988 to January 2012. The sample period is determined by the
availability of bankruptcy data. 59,384 loans were originated over the sample period, with
inflation adjusted aggregate dollar amount of $15,065 billion. The number of loans increases
over years, and decreases following the financial crisis, reflecting the tightening of credit
markets. Banks are more cautious when issuing loans following the crisis (Murfin, 2012). The
first cluster of default and bankruptcy announcements is in the period 1999 to 2003, and the
second is during the recent financial crisis. The increased number of bad loans leads to more
affected entities, both banks and borrowers. A total of 1,108 firms either default or file for
Chapter 11 bankruptcy (Column 3). When the firm defaults first and then files for bankruptcy, it
enters into the summary statistics once at the original default date. The number of bad loans
exceeds the number of borrowers because many distressed borrowers have multiple loans
12
outstanding (Column 4). The number of affected banks is much greater than what has been
documented in the literature with the inclusion of syndicate participants (Column 6). The number
of total lenders on each loan varies, but most are over three. The last column presents the total
number of borrowers that are affected by the distress news. Each affected bank or borrower only
enters into the summary once a year even if it is affected multiple times.
Insert Table II here.
Table II Panel B presents a brief summary of the accounting information of the entities
examined in the study. There is a large variation in the size of the distress borrowers. The
smallest 25% have $250 million or less in total assets, and the top 25% has over $1.7 billion in
total assets. Net income is negative across the distress sample, indicative of troubled financial
conditions. The distress firms are generally highly levered, and are often short in cash; showing
difficulty with paying back the outstanding loans. When something goes wrong in their daily
operation, these firms are less likely to recover and more likely to default or perhaps go
bankrupt.
Comparing to the distressed borrowers, the banks are generally much larger. Bank recovery
of bad loans is low, only about 20% of the loan charge off. This suggests a bad loan is likely to
cause a permanent damage to the bank. This agrees with the argument that bad loans can and
should have material damage to banks’ expected value, especially when they are relatively large
and come in clusters. To avoid potential chaos and better shield them from troubled loans, banks
are required maintain some loan loss allowance. The mean loan loss allowance is approximately
2% of banks’ total assets. The amount of loss allowance is constantly changing depending on
banks’ expectation about the performances of their loans. When the market becomes more
volatile, or when the banks are expected to have more bad loans in their portfolio, they are likely
13
to adjust the loan loss allowance number upward, so the impact to its real operation can be
managed. In addition, when a bank is larger in size, its loan loss allowance is greater, so it is able
to absorb bigger loss on loans. As a result, the impact of bad loans should be smaller.
The affected borrowers share the same lender with the default or bankrupt firms, but unlike
the distressed firms, their financial positions are sound and promising. The primary focus of the
paper is on the publicly traded affected borrowers, which are relatively large and transparent.
Median market value of these firms is $1.6 billion. The bottom 25% have less than $500 million
in total assets, while the top 25% have over $4.6 billion. Unlike the distressed borrowers, the
affected borrowers’ net income is usually positive with a mean of $95 million, they tend to have
more cash holding than the financially troubled firms.
4. Empirical results
4.1 Univariate tests of CARs
Table III summarizes the average cumulative abnormal return using Fama-French-Carhart
four-factor model of the distressed borrowers, affected banks, and affected borrowers over
different windows surrounding the default or bankruptcy announcement date. Not surprisingly,
the equally-weighted four-factor adjusted abnormal return for the distress firm is -8.41% at the
announcement day, which is highly significant (Panel A). Over the longer seven-day event
window, the cumulative abnormal return is more negative, a stunning -20.58%. Distress firms
lose more than one fifth of their value due to the distress news. The loss is greater when it is a
bankruptcy filing instead of default announcement, which suggests market is still unclear about
the firm’s future outlook when it defaults on its debt obligation, but once it files for bankruptcy,
such expectation fades away. The difference between the two announcements’ impact is only
marginal significant in most event-windows.
14
Insert Table III here
Panel B presents the four-factor adjusted abnormal return of the affected banks. Banks, in
general, experience negative stock return during the distress announcement. Although the event-
day abnormal return is only marginally significant, the cumulative abnormal returns using other
event-windows are all highly significant. The average seven-day bank CAR is -0.8%, and it is -
1.11% when using the eleven-day window, which spans from 5 days prior to the announcement
and 5 days afterwards. The impact of the distress news is long-standing. Opposite from the
distressed borrowers, which experience more negative returns for bankruptcy filing, the banks
experience more negative returns for defaults than bankruptcies. This is possibly due to the fact
that default often precedes bankruptcy, so it is a big surprise to the market. Similar results are
documented in the literature (Dahiya et al. 2003). However, more importantly, regardless of the
role played by the banks in the bad loans, lead arrangers or syndicate participants, the CARs are
all negative. Unexpectedly, the CARs of syndicate participants are actually more negative than
the lead, especially for the announcement day return. The difference is both statistically and
economically significant. Moreover, when dividing the banks into quintiles from highest to
lowest based on their exposure to the bad loans, the banks with the highest exposure to bad loans
perform significantly worse than the ones that has the lowest exposure to bad loans. The average
difference over the 11-day window is approximately -3.12%, which is statistically significant at
all confidence levels. It is evident that the degree of impact to each lender depends heavily on the
level of exposure it has toward the bad loan. When a bank fails to properly manage its exposure
to any individual loans, it is likely to experience a bigger loss when the loan goes bad.
When the impact is economically significant at the bank, its relationship borrowers are
adversely affected even if the shocks are not directly related to them. The negative shocks
15
transmitted via the lending channel through the common lender when the lender is unable to
manage the damage at the bank level. Figure 1 illustrates the average return of the affected firms
appears to follow a random walk process prior to and after the distress announcement. There is,
however, a significant drop during the announcements of either default or bankruptcy.
[Insert Figure 1 here]
The four-factor adjusted abnormal returns for the web borrowers are significantly negative.
Similar to the impact on lenders, the default announcement affects their stock return more than
the bankruptcy news. The mean difference over the seven-day event window is -0.4%, which is
highly significant, both statistically and economically (Panel C). This suggests more information
is contained in default announcements because it is a bigger surprise to the investors. In addition,
opposite from the common belief that borrowers should experience greater value loss when their
lead arrangers suffer reputation damage, the test results reveal that borrowers’ losses are worse
when their lead arrangers are loan syndicate participants. Mean differences between the two
types of affected borrowers using multiple event-windows are marginally significant. However,
such outcomes could be attributed to the fundamental difference between the syndicate
participants and lead arrangers. In general, the lead arrangers are more reputable and larger in
size, which is less likely to be affected by some small and isolated bad loans. In addition, the
web borrowers perform significantly worse than their competitors that are in the same 4-digit
SIC industry. It suggests the negative impact from the lending channel can potentially jeopardize
firm’s position within the industry.
The results of the univariate test indicate both affected banks and borrowers experience
material loss due to distress announcement, but the test does not control for other factors that
could jointly influence stock returns, such as other borrower or bank specific characteristics. The
16
need for better understanding what drives the stock return requires more comprehensive
multivariate cross-section tests.
4.2 Cross-section analysis of banks’ CARs
The first analysis examines determinants of bank stock performance over the distress
announcement. In particular, how default or bankruptcy news impacts banks’ stock return, what
factors exacerbate the influence, and what characteristics alleviate the impact. A bank that is
more exposed to the distressed borrower should be more affected by the news. Lead arrangers
should have more negative stock return because they experience both reputation and capital
damage.
To test the predictions, I estimate the following regression:
where 𝐿𝑎𝑟𝑔𝑒 𝑑𝑎𝑚𝑎𝑔𝑒𝑗,𝑡−1is a dummy variable that equals to 1 when the aggregate default damage experience by bank j at year t-1 is in the top quintile of distress
damage after ranking it from high to low, and 0 otherwise. X, Y, and Z represent a series of control variables that measure borrower, bank, and loan’s characteristics,
respectively. Detailed definitions of the control variables are given in Appendix. The alternative measure of distressed loan damage is a continuous variable that
aggregates the amount of all outstanding distressed loans. The sub-sample analysis excludes loans originated during the crisis periods. Heteroskedasticity-consistent p-
values clustered at the firm level are reported in parentheses. The superscripts ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.