The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Fourteenth
International
Banking
Conference
Federal Reserve Bank
of Chicago
First Quarter and Second Quarter 2011
~
research library
Federal Reserve Bank
of St. Louis
JUN 2 3 2011
Economic.
perspectives
First Quarter
2
Competition in mortgage markets: The effect of lender
type on loan characteristics
Richard J. Rosen
22
Monitoring financial stability: A financial conditions
index approach
Scott Brave and R. Andrew Butters
Second Quarter
44
Understanding the Great Trade Collapse of 2008-09
and the subsequent trade recovery
Meredith A. Crowley andXi Luo
71
How do private firms use credit lines?
Sumit Agarwal, Souphala Chomsisengphet, and John C. Driscoll
Economic
perspectives
President
Charles L. Evans
Executive Vice President and Director of Research
Daniel G. Sullivan
Senior Vice President and Economic Advisor
Spencer Krane
Senior Vice President, Financial Markets Group
David Marshall
Microeconomic Policy Research
Daniel Aaronson, Vice President
Macroeconomic Policy Research
Jonas D. M. Fisher, Vice President
Markets Team
Richard Heckinger, Vice President
Finance Team
Anna L. Paulson, Vice President
Regional Programs
William A. Testa, Vice President
Economics Editor
Lisa Barrow, Senior Economist
Editors
Helen O’D. Koshy
Han Y. Choi
Graphics and Layout
Rita Molloy
Proofreading
Sheila A. Mangier
Production
Julia Baker
Economic Perspectives is published by the Economic
Research Department of the Federal Reserve Bank of Chicago.
The views expressed are the authors’ and do not necessarily
reflect the views of the Federal Reserve Bank of Chicago
or the Federal Reserve System.
© 2011 Federal Reserve Bank of Chicago
Economic Perspectives articles may be reproduced
in whole or in part, provided the articles are not reproduced
or distributed for commercial gain and provided the source
is appropriately credited. Prior written permission must
be obtained for any other reproduction, distribution,
republication, or creation of derivative works of Economic
Perspectives articles. To request permission, please
contact Helen Koshy, senior editor, at 312-322-5830 or email
Helen.Koshy@chi.frb.org.
Economic Perspectives and other Bank
publications are available at www.chicagofed.org.
fi? chicagofed. org
ISSN 0164-0682
Contents
First and Second Quarters 2011, Volume XXXV, Issues 1 and 2
First Quarter
2
Competition in mortgage markets: The effect of lender
type on loan characteristics
Richard J. Rosen
This article examines how competition among lenders affects mortgage loan characteristics.
The author finds that, on average, banks issue safer mortgages than independent mortgage banks.
Further, mortgages from banks with a branch in the local market where the property is tend to be
safer than mortgages from banks without a local branch. Changes in market shares among lender
types (local bank, nonlocal bank, or independent mortgage bank) that lead to higher loan risk also
are associated with better borrower quality. Increasing the local market share of a lender type raises
loan risk and borrower quality at that lender type.
22
Monitoring financial stability: A financial conditions
index approach
Scott Brave and R. Andrew Butters
Monitoring financial stability requires an understanding of both how traditional and evolving
financial markets relate to each other and how they relate to economic conditions. This article
describes two new indexes of financial conditions that aim to quantify these relationships.
Second Quarter
44
Understanding the Great Trade Collapse of 2008-09
and the subsequent trade recovery
Meredith A. Crowley and Xi Luo
This article documents the Great Trade Collapse of 2008-09, as well as the dramatic recovery in
trade of 2009-10. The authors consider how three distinct policy actions—fiscal stimulus, funding
for trade finance, and a commitment to refrain from increasing trade barriers—might have affected
both the collapse and recovery.
71
How do private firms use credit lines?
Sumit Agarwal, Souphala Chomsisengphet, and John C. Driscoll
The authors find that firms that face higher upfront commitment fees, risk premium spreads, or usage
fees have smaller credit lines, while those with higher overdraft fees have larger ones. Firms with
greater profit growth in the past have larger credit lines, while those with more internal funds or higher
volatility in profit growth have smaller credit lines. The results for line utilization are quite similar.
80
International Banking Conference
The Role of Central Banks in Financial Stability: How Has It Changed?
Competition in mortgage markets: The effect of lender type
on loan characteristics
Richard J. Rosen
Introduction and summary
The years 1995 through 2007 saw a boom and bust in
home prices and purchase activity in the United States.
There has been a lot of attention paid to the causes of
the boom-bust cycle and who, or what, is to blame.1
Some have blamed the cycle on subprime lending and
the securitization of home mortgages (see, for example,
Mian and Sufi, 2009; Keys et al., 2010; and Demyanyk
and Van Hemert, 2009).2 During the latter years of the
boom, both subprime lending and securitization ex
panded significantly. By 2005, subprime lending was
over six times as large as its pre-2000 peak, and over
all securitization was more than twice its pre-2000 peak?
But these changes, and the housing cycle in general,
were not uniform across the country. The expansion of
lending and the subsequent problems in housing mar
kets were more extreme in some markets than in others
(Mian and Sufi, 2009), in part possibly because of
changes in home prices. Home prices rose much more
rapidly in some markets than in others, both in percent
age terms and relative to fundamentals (see, for example,
Haines and Rosen, 2007). Differences across markets
may occur because of market conditions and the core
attractiveness of a market (see, for example, Gyourko,
Mayer, and Sinai, 2006). However, they may also
reflect differences in the composition of lenders in par
ticular markets. This article explores how the charac
teristics of mortgages varied over time and across
markets and how these differences relate to the com
position of lenders in the markets.4 The characteristics
1 focus on are measures of loan risk and borrower qual
ity. 1 examine how these differ across mortgages issued
by different types of lenders and how shifts in mortgage
shares among lender types in local markets affected
standards of lenders in those markets.5
I focus on the lender that originates, or originally
funds, a mortgage. The primary division of lenders
is into banks (that is, depository institutions) and
2
independent mortgage banks (IMBs). Banks and IMBs
differ in corporate strategy and regulation, both of
which may affect their approach to participating in
mortgage lending, including the characteristics of the
mortgages they issue and the borrowers they issue
them to. Mortgage lending generally plays a much
larger role at IMBs than at banks; unlike IMBs, many
banks tend to view mortgages as just one part of a
broader strategy. Banks typically have branch networks
to attract deposit customers, and mortgages may form
only a part of their asset portfolios. In part because the
presence of branches can affect the way banks com
pete for mortgage borrowers, I subdivide banks by
whether or not they have branches in the local market
being considered (local banks versus nonlocal banks).
Local banks may be able to use their branches’ pres
ence to help them capture potential borrowers. Over
the past 15 years, the market shares of the three types
of lenders (local banks, nonlocal banks, and IMBs)
have shifted by as much as 15 percentage points. From
1995 through mid-2006, the share of mortgages made
by local banks trended down. Initially, local banks lost
market share to IMBs, but starting in 2001, mortgages
issued by nonlocal banks began to make up a large
share of total mortgages in many markets. Finally,
there was a massive readjustment away from mortgages
made by IMBs starting in mid-2006, slightly after the
housing market bust had begun.6
The way I divide lenders in this article reflects
important differences across lenders in the mortgage
delivery process. How borrowers are matched with
lenders and how mortgages are ultimately financed
Richard J. Rosen is a senior economist and economic advisor
in the Economic Research Department at the Federal Reserve
Bank ofChicago. He thanks GeneAmromin and Anna Paulson
for their comments and Robert McMenamin and Edward
Zhongfor their assistance with the research.
1Q/2011, Economic Perspectives
(two key elements) typically differ across the three types
of lenders I focus on. A potential borrower wanting a
mortgage has the option of contacting a bank or IMB
directly. For example, a borrower who wants to find
out about lending terms and conditions could visit local
bank branches and talk with a loan officer. Alterna
tively, the borrower could use the services of a mort
gage broker. A mortgage broker is an independent
agent who serves as a contact between borrowers and
lenders, arranging loans but not actually lending money.
The broker can offer borrowers a menu of loan products
from different lenders.7 According to one study, mort
gage brokers helped arrange 68 percent of all residen
tial mortgages in 2004.8 Brokers make it easier and less
expensive for lenders with no physical presence in a
market to lend in the market. This can potentially help
both banks and IMBs expand. Often, the use of brokers
is referred to as wholesale lending (as opposed to retail
lending, where originators connect directly with borrow
ers, often when customers visit a bank branch or have
a pre-existing relationship with the lender).9 The expec
tation is that most IMBs and nonlocal banks operate
in the wholesale lending market, while local banks
rely on a mix of retail and wholesale lending (although,
clearly, there are variations in strategy across banks
of the same type).
As noted previously, many loans are securitized.
Traditionally, the primary option for a potential home
purchaser who needed a loan was to go to a local bank.
Typically, the bank would hold the loan in its asset port
folio, financing it using its own deposits. This put a
natural limit on the ability of the bank to issue mort
gages. In the securitization process, the bank or other
lender that initially funds the loan quickly sells it to a
third party. The third party then uses a pool of mort
gages as the collateral backing a bond issue. The bonds,
known as mortgage-backed securities, are sold to in
vestors (see Rosen, 2007b). The ability to easily sell
mortgages means that the originating lender can finance
a larger volume of loans with its capital. The costs and
risks of originating mortgages for lenders that plan to
securitize them are different than for lenders that plan
to keep the loans in their portfolios. This difference may
affect how the lenders compete for borrowers. While
securitization made it easier for all lenders to expand,
it is likely to be more important for those lenders with
out a strong deposit base, especially IMBs.
The ties between mortgage brokerage and securi
tization, on the one hand, and lender competition and
lending market standards, on the other hand, are both
direct and indirect. The presence of mortgage brokers,
at least those who act in the interests of the home buyers
(see note 7), should increase the competitiveness of
Federal Reserve Bank of Chicago
lenders. This could mean lower mortgage rates, but it
also could mean that other mortgage terms are relaxed,
such as allowing applicants to take out larger mortgages
than their incomes might readily support or mortgages
that are significantly higher than the value of the homes
they are buying. It is plausible that increased compe
tition among lenders contributed to such developments
as the 125 percent loan-to-value mortgages offered
during the housing boom. Securitization also can in
crease the competition for mortgages. The expansion
of securitization in the 1990s and the early part of the
2000s meant that the risk that a lender would not be
able to sell a loan was reduced; also, the time a lender
was forced to hold the loan before selling it as part of
the securitization process likely fell. This made it less
risky, and therefore less expensive, for lenders to enter
new markets and expand. However, securitization also
benefits from economies of scale. This led to industry
consolidation. In 1995, the ten largest mortgage origi
nators made 25.3 percent of all mortgages; by 2005, it
was 32.7 percent.10 Thus, the net impact of securitiza
tion on lender competitiveness is unclear.
It is likely that the mortgage delivery system, in
cluding the use of brokers on the front end and securiti
zation on the back end, affects how lenders compete,
including how lending market standards are set. How
ever, the lack of data makes it difficult to directly tie
brokerages and the rest of the mortgage delivery system
to market conditions. The primary data on mortgages
come from the information lenders are required to re
port to the Federal Financial Institutions Examination
Council under the Home Mortgage Disclosure Act
(HMDA). The HMDA data identify lenders and give
some information on the disposition of a mortgage,
but they do not include information on how a mortgage
applicant connects with a lender, including whether a
broker was involved in the lending process. The sup
plementary data on mortgages that I use in this article—
from Lender Processing Services (LPS) Applied
Analytics (formerly known as McDash Analytics)—
also do not have information on the front end of the
mortgage process. The best option I have is to use
information on lenders as a proxy for the mortgage
origination processes they use—and thus the lenders’
effect on lending market competition and conditions.
I use HMDA and LPS data to examine both how
mortgage characteristics differ by lender type and how
the distribution of lender types within a market affects
mortgage characteristics in the market. I find that, on
average, banks make ex ante safer loans than IMBs
do, both on an absolute scale and relative to IMBs in
the counties where they lend. Also, mortgages issued
by banks have lower loan-to-income ratios and lower
3
loan-to-value ratios, and banks’ borrowers have high
er FICO (Fair Isaac Corporation) scores.11 Among
banks, I find that local banks make safer loans than
nonlocal banks do, with nonlocal banks falling be
tween local banks and IMBs.
I examine how the shift in lending in a market from
one type of lender to another affects all the lenders in
a market. This gives an indication of whether lender
type affects how a firm competes. If lender type does
not matter, then the shift in lending should have no
impact. I find that a shift in lending toward a particu
lar type of lender is associated with a larger change in
lending standards at that type of lender than at other
types of lenders. The interesting thing is that when a
particular category of lender increases its share in a
local mortgage market, that category of lender makes
mortgages with higher loan risk, but to borrowers who
are, on average, of higher quality. For example, when
the mortgage share of local banks in a market increases,
those banks issue mortgages with higher loan-to-income
and loan-to-value ratios (higher loan risk), but to bor
rowers with higher FICO scores (lower borrower risk).
The impact of a change in the share of mortgages issued
by a particular type of lender on other types of lend
ers is much weaker. So, for example, a shift in the
share of mortgages issued from local banks to IMBs
has a generally insignificant impact on loan standards
at nonlocal banks.
I also examine whether large metropolitan areas
are different from less densely populated areas. Sepa
rating counties (markets) into those in large metro
politan statistical areas (MSAs) and those in small
MSAs,121 find that the impact of an increase in the share
of a particular category of lender on that category’s
lending standards is weaker in the large-MSA counties
than in the small-MSA counties.
Data
The primary source of mortgage data that I use
comes from information that lenders are required to
report under the Home Mortgage Disclosure Act. HMDA
mandates that lenders report data for the vast majority
of mortgage applications.13 For each application, the
HMDA data provide the name of the lender, its type,
and loan information, including the location of the
borrower. Lenders are required to report information
on all types of residential mortgages, including loans
used for purchases of single-family homes, loans used
for purchases of multifamily dwellings, loans to refi
nance existing mortgages, and loans for home im
provement. To make the comparisons in this article
as revealing as possible, I restrict the sample to loans
used for purchases of single-family homes and, within
4
single-family loans, drop both second mortgages and
home equity lines.14 For most of the analysis, I separate
lenders by whether or not they also take deposits. In
stitutions that both make loans and take deposits are
regulated and chartered differently from those that
only make loans. The deposit-taking institutions, which
I generically refer to as banks, comprise commercial
banks, thrift banks, and credit unions.151 refer to the
non-deposit-taking lenders as independent mortgage
banks, and this category includes specialized mortgage
lenders and independent finance companies.
One important drawback of the HMDA data is that
a lender is classified without regard for whether the lender
is the subsidiary of a different kind of institution. So,
a mortgage made by a mortgage bank that is the sub
sidiary of a commercial bank holding company is classi
fied by HMDA in the IMB category. Instead, I classify
lenders by the type of lender that their parent organi
zation is. This assumes that major strategic choices are
made at the parent organization level. This also assumes
that where a lender books a mortgage is a matter of
lender policy, meaning, for example, that some parent
organizations book these loans at a bank subsidiary, while
others book them at a mortgage bank subsidiary.16
In this article, I use quarterly HMDA data from
1995 through 2007. During this period, total mortgages
issued increased from 1995 through the third quarter
of 2005 (see figure 1). However, the rate of increase
was not constant. From 1995 through 1999 (the early
run-up period), home purchases increase at a rate of
8.4 percent. This falls to a rate of 3.8 percent from 2000
through 2003 (the mid run-up period), before rocket
ing up at a rate of 11.9 percent from 2004 through the
third quarter of 2005 (the late run-up period). From
the fourth quarter of 2005 through 2007 (the housing
bust), there is a sharp decline in home purchases. The
pattern is superficially similar to the pattern in home
prices, as indicated by the Federal Housing Finance
Agency’s (FHFA) House Price Index (HPI), also re
ported in figure l.17 But home prices increased faster
during the 2000-03 period than during the 1995-99
period (see, for example, Haines and Rosen, 2007,
for a discussion of home price changes).
There is likely to be a difference in how banks
connect with potential borrowers, depending on their
presence in a market. Potential borrowers connect with
a bank because of a pre-existing relationship, such as
a checking or savings account. They may also walk
into (or phone) one of the bank’s branches. These two
approaches are likely to be correlated with the bank
having a physical presence (that is, a branch) in the
borrower’s local market. I define a mortgage as com
ing from a local bank if the lending bank has a branch
1Q/2011, Economic Perspectives
FIGURE 1
Total number of mortgages and home prices
millions of mortgages
index, 1992 = 100
3.0
Mid run-up period
Early run-up period
2.5
Late
run-up
period
300
Housing
bust
250
2.0
200
1.5
150
1.0
100
0.5
50
0
0
1995
’97
’99
2001
’03
-------
Mortgages (left-hand scale)
-------
FHFA HPI (right-hand scale)
’05
’07
Note: See the text for details on the four periods.
Sources: Author’s calculations based on data from the Home Mortgage Disclosure
Act; Federal Deposit Insurance Corporation, Summary of Deposits; and Federal
Housing Finance Agency (FHFA), seasonally adjusted purchase-only House Price
Index (HPI), from Haver Analytics.
Federal Reserve Bank of Chicago
in the county where the home pur
chased with the mortgage is located.
Alternatively, a borrower may use
a mortgage broker (or an Internet
equivalent) to help choose a lender.
Brokers allow a bank to make mort
gages without having a physical
presence to attract customers. I de
fine a mortgage as coming from a
nonlocal bank if the lending bank
has no branches in the county where
the home purchased with the mort
gage is located. While I do not know
whether a borrower has a pre-exist
ing relationship with a bank, walks
into a branch, or uses a broker, I as
sume that it is more likely that a loan
from a local bank is made through
a branch or pre-existing relationship
(that is, the retail channel). The vast
majority of loans made by nonlocal
banks (and IMBs) come through
brokers (that is, the wholesale chan
nel). In the entire sample, 28.46
percent of mortgages are made by
local banks and 40.45 percent are
made by nonlocal banks (of course,
a bank can be a local bank in some
markets and a nonlocal bank in other
markets).18
Figure 2 shows the share of
mortgages made by local banks,
nonlocal banks, and IMBs over the
sample period. The share of mort
gages made by local banks declined
steadily from 1995 through the third
quarter of 2006, that is, during the
period when housing prices rose and
into the start of the housing bust.
In the first quarter of 1995, local
banks had a share of 34.26 percent
of the mortgages made, but by mid2006, this share had decreased to
23.84 percent. In 1995-99 (the early
run-up period), the drop in the num
ber of mortgages made by local
banks was balanced by the rise in
the number of mortgages made by
nonlocal banks. But as home prices
began to increase at a faster pace,
the share of mortgages made by IMBs
began to rise. At the start of 2000,
IMBs had a share of 26.68 percent
5
of the mortgages made, but this
FIGURE 3
quickly increased to 37.10 percent
Mortgage application denial rates, by lender type
in the first quarter of 2005. Starting
in late 2005, as home prices began
to fall and private securitization
markets shut down, these patterns
reversed. By the end of 2007, the
share of mortgages made by local
banks increased to 37.90 percent,
while the share of mortgages made
by IMBs fell to 21.59 percent. Note
that the decline in IMB share in
2006-07 is at least partially due to
the failure of American Home
Mortgage and several other IMBs.
Up to now, I have been exam
ining mortgages issued by lenders.
But HMDA data also include re
-------- Local bank
cords for mortgage applications
------- Nonlocal bank
that are denied. One focus of this
------- Independent mortgage bank
article is to examine how lender
competition affects the characteris
Note: See the text tor details on the four periods.
Sources: Author’s calculations based on data from the Home Mortgage Disclosure
tics of loans that are made. For the
Act; Robert Avery, Board of Governors of the Federal Reserve System; and Federal
most part, I treat the denial rate as
Deposit Insurance Corporation, Summary of Deposits.
if it is a loan characteristic, viewing
it as a signal of the aggregate riski
ness of loans that are granted. A
lower denial rate may mean higher loan or borrower
mortgages that are granted, I need additional data.
risk. To the extent that we do not perfectly observe
The HMDA data include information on the amount
loan and borrower risks, the denial rate can serve as a
of each loan and the income of the borrower that I use
proxy for them. Figure 3 reports the percentage of mort
to get the ratio of loan amount to income. However,
gage applications that are denied by lender type.19 The
to go further, I incorporate data from another source.
mortgage denial rate of local banks was flat for most
As I mentioned before, to supplement the HMDA
of the sample period, only showing the beginning of
data, I get information on loan details and borrower
an increase when home prices fell toward the end of
quality from LPS Applied Analytics, which collects
the sample. The mortgage denial rate of nonlocal banks
data from a number of large loan servicers. These
dropped sharply as home prices began to rise more
data include detailed information on mortgage char
quickly in 2000: The denial rate fell from 40.22 percent
acteristics and payments, as well as on the borrower.
in the second quarter of 2000 to 13.44 percent in the
The LPS data contain information on the mortgage at
second quarter of 2002. The denial rate of nonlocal
origination and a monthly record of its status. I match
banks then drifted up to about 25 percent by the end of
the LPS data to the HMDA data. Because of data lim
the sample, in 2007. IMBs followed a similar pattern
itations, it is not possible to match an LPS observation
to that of nonlocal banks, perhaps because both groups
with each HMDA record. The final merged data set
are wholesale lenders, getting most of their loans from
matches 38.6 percent of the LPS records and 18.4 percent
mortgage brokers. As I noted before, while local banks
of the HMDA records. The matched records are broadly
may get some applicants through brokers, they can also
representative of the LPS sample. The proportion of
appeal to people with whom they have a pre-existing
different lender types is similar, as is the mean loanrelationship or to people who visit a local branch.
to-income ratio. However, the merged data underrep
The differences in mortgage denial rates across
resent certain loans in the HMDA data. Because LPS
lenders, and possibly across time, likely reflect in part
Applied Analytics only gets data from a limited number
differences in applicant quality. They may also result
of large servicers, it misses many loans kept in portfolio
from variation in the types of mortgages that applicants
by smaller banks or serviced by smaller servicers.
want. To examine whether these differences affect the
The LPS data also underrepresent subprime loans
6
1Q/2011, Economic Perspectives
(see the discussion later on this). Finally, the share
of loans in HMDA data that are matched to LPS
data increases over the sample period, paralleling
the increased servicer coverage by LPS.
Differences across lender types
In this section, I present information on how
various loan and borrower characteristics differ across
lender types. Again, I focus on three lender types: local
banks, nonlocal banks, and IMBs. Banks include all
depository institutions (commercial banks, thrift banks,
and credit unions); when appropriate, I discuss the
different depository institutions.
The differences in mortgage characteristics across
lender types are presented in three different ways in
table 1. Panel A of table 1 presents full sample means.
For each variable, I take the mean for each quarter of
the sample period. The mean and standard deviation
of the quarterly means are reported in panel A. I take
the mean of quarterly means rather than the mean of the
entire sample because the number of loans increases
over time, and I do not want the means to overweight
the latter part of the sample. One issue with using these
means to compare lender types is that lender types are
not uniformly distributed across markets. As a control
for this, I take the average of each variable for each local
market in each quarter, using counties as local markets.
Panel B of table 1 reports the average difference be
tween the local market average for a lender type and that
market’s average for all lenders. This is informative about
how loans differ across lender types. For example, the
proportion of fixed-rate mortgages (FRMs)20 at local
banks is 77.17 percent, which is 0.29 percentage points
lower than the average proportion of fixed-rate mortgages
at all lenders (seventh row in panel A). Does this mean
that local banks give too few fixed-rate mortgages? Not
necessarily. As shown in panel B (sixth row), local banks
give 2.45 percentage points more fixed-rate mortgages
than the average of lenders in the markets they are in.
This suggests that lenders in markets with many local
banks issued a smaller percentage of fixed-rate mortgages
than did lenders in other markets. Finally, as figure 1
(p. 5) shows, the sample period includes a period of in
creasing sales and prices followed by a period of declin
ing sales and prices. Rather than chart every variable,
I report the sample averages for three interesting quarters
in panel C of table 1.1 show the values in the first period
a variable is in the sample; the fourth quarter of 2004,
to reflect the peak of sales and prices; and the fourth
quarter of 2007, to observe the effects of the declining
sales and prices. In general, the three different ways
of looking at the data indicate the same patterns, but
I discuss them in more detail when they do not.
Federal Reserve Bank of Chicago
I use the data in table 1 to examine how mort
gage characteristics differ by lender type. In doing
this, it is useful to divide mortgage characteristics
roughly into three groups. The first is loan risk. These
are the features that have to do with risk introduced
by the size of the mortgage. The second is borrower
quality. These characteristics measure the risk of the
borrower more than the mortgage itself. There will be
some overlap in the first two groups. Finally, I include
some variables that are likely to be more weakly cor
related to loan or borrower risk.
The first characteristic is the ratio of the loan amount
to the borrower’s income. Borrowers with a larger loan
relative to income, all else being equal, are more like
ly to have trouble paying their mortgages. To measure
the loan-to-income ratio, I divide the amount of the loan
by the borrower’s reported income from the HMDA
data.21 Figure 4 (p. 10) charts this ratio for the three
types of lenders over the sample period. Several things
are apparent from the data. On average, IMBs lend more
per dollar of income than banks do (see also table 1,
panel A, second row). While not shown in the figure,
mortgages issued by thrift banks have a higher average
loan-to-income ratio than do mortgages issued by com
mercial banks, and the mortgages made by credit unions
have the lowest ratio of all lender types. The raw aver
ages across the types of banks (table 1, panel A, sec
ond row) indicate that local and nonlocal banks lend
the same amount as a fraction of borrower income—
that is, 2.31. But, mortgages issued by local banks have
a loan-to-income ratio (table 1, panel B, first row) that
is 0.07 (7 percentage points) lower than the average
of lenders in the markets they are in, while mortgages
from nonlocal banks have a ratio that is only 0.02 lower,
with the difference between 0.07 and 0.02 being statis
tically significant. This would arise if local banks had
made a lot of mortgages in markets where the loan-toincome ratio was higher than in those markets where
nonlocal banks made a lot of mortgages, so that the
2.31 loan-to-income ratio for local banks is 0.07 below
the average of lenders in their markets, while the 2.31
ratio for nonlocal banks is only 0.02 below the average
of lenders in their markets. The loan-to-income ratio for
all lenders rose significantly over the sample period,
from 2.08 in the first quarter of 1995 to 2.61 in the
last quarter of 2007 (table 1, panel C, second row).
The rate of increase of the loan-to-income ratio was
fastest from 2000 through 2004, precisely when home
prices were rising most quickly (see figure 4, p. 10).
A second measure of loan risk is the loan-tovalue ratio (table 1, panel A, third row), available
from the LPS data. This is the ratio of the mortgage
amount to the appraised value of the home.22 The
7
oo
TABLE 1
Summary statistics, by lender type
A. Means
Independent
All lenders
Mean
Local banks
Standard
deviation
2.35
83.07
Standard
deviation
Mean
Standard
deviation
mortgage banks
Mean
Standard
deviation
28.46
3.12
40.45
2.06
31.10
3.56
0.18
2.38
2.31
81.03
0.22
2.32
2.31
83.21
0.16
2.44
2.42
84.72
0.20
2.14
Lender share
Loan-to-income ratio
Loan-to-value ratio
Mean
Nonlocal banks
FICO score
707.6
5.09
715.7
5.14
706.0
6.11
699.2
10.18
Loan denial rate
24.23
0.060
13.47
0.018
26.97
0.096
28.06
0.062
Subprime share
Fixed-rate mortgage share
2.61
77.46
2.84
11.61
2.36
77.17
2.81
10.90
2.99
76.44
3.22
12.77
2.29
79.32
2.66
11.13
Jumbo share
8.20
2.79
10.22
3.94
8.47
2.46
5.87
2.29
Portfolio share
8.22
2.76
14.09
5.75
7.23
2.73
3.81
2.34
23.20
68.58
7.00
6.75
20.10
65.81
6.65
8.36
19.66
73.11
7.69
8.05
28.91
67.29
9.88
10.08
Private share
Government share
Unemployment rate
Income per capita
4.72
0.37
35,329
1,450
Independent
B. Within-county differences
Local banks
Mean
Nonlocal banks
Standard
deviation
Mean
Standard
deviation
mortgage banks
Mean
Standard
deviation
1Q /2 011, Econom ic Per spe ctiv es
Loan-to-income ratio
-0.07
0.03
-0.02
0.03
0.07
0.05
Loan-to-value ratio
FICO score
-2.44
8.45
0.47
3.61
-0.05
-1.12
0.32
2.16
1.39
-7.55
0.77
7.01
Loan denial rate
-5.41
0.96
0.16
1.66
7.04
1.41
Subprime share
-0.81
1.07
0.76
0.99
-0.22
1.20
2.45
0.53
2.68
0.005
-1.50
0.32
1.86
0.002
0.10
-0.82
1.50
0.004
2.50
Fixed-rate mortgage share
Jumbo share
Portfolio share
4.49
3.47
-0.27
1.54
-3.36
Private share
-1.95
2.89
-2.98
4.63
8.03
7.00
Government share
-3.33
3.91
3.37
4.67
-4.12
7.60
Federal Reserve Bank of Chicago
is fo r 1997 :Q1 .
30.03
60.36
9.61
—
2.08
87.22
698.46
27.58
0.09
61.50
4.87
—
81.96
715.40
22.55
0.66
89.19
5.61
7.75
8.24
84.00
—
2.60
79.46
709.63
18.36
8.30
56.59
14.21
9.37
33.17
57.46
2.61
2007 :Q4
2004: Q4
All lenders
59.22
6.18
18.68
23.40
57.92
0.11
34.25
1.99
85.36
712 .57
15.33
1995 :Q1 a
7.55
55.14
18.26
16.25
33.05
50.70
13.31
77.33
717.13
2.61
26.22
2004 :Q4
Local bank s
78.68
6.11
15.21
37.90
2.59
79.34
723.62
18.55
0.25
87.07
8.70
2007 :Q4
0.04
59.66
6.05
9.69
22.69
67.62
28.31
2.51
2.10
87.42
703.72
13.27
7.55
33.95
58.50
55.31
79.49
709.76
17.90
8.55
40.99
36.21
40.52
2.55
82.79
712 .78
23.5 7
0.98
88.60
4.97
4.43
7.74
87.83
3.49
5.77
37.35
56.89
0.11
63.81
87.97
666.83
37.29
2.11
29.54
1995 :Q1 a
4.76
32.12
63.12
11.31
60.13
8.71
700.39
21.98
81.71
32.79
2.74
1.52
1.37
12.92
85.71
21.59
2.75
84.89
705.23
26.64
0.74
94.00
2007 :Q4
200 4:Q 4
2007 :Q4
200 4:Q 4
1995 :Q1 a
Inde pen den t mortgag e banks
Nonlo cal banks
Notes: All valu es are in perc ent exce pt thos e for loan -to-inco me ratio; FICO scor e, which indic ates the Fair Isaac Corpora tion cred it scor e; and inco me per capita, whic h is in dollars. Full def initions for the variable s
are in the text. Where possible, the statisti cs in the "all lend ers ” colu mns are for all Home Mortg age Disc losu re Act (HMD A) obs erva tions, while all other statisti cs are for obs erva tions in the Fede ral Rese rve Ba nk of
Chicag o data set tha t merg es the HMD A data and the Len der Proc essing Serv ices (LPS) App lied Ana lytic s data. The means and stan dard dev iatio ns are deriv ed from the average of qua rter ly means in local marke ts
for the period 19 95 -20 07 (excep t for FICO score, whic h starts in 1997). Panel B report s the average difference betw een the local market ave rage for a lend er type and tha t ma rke t’s ave rage for all lende rs. In panels
A and C, cert ain sha res ma y not total beca use of roundin g.
Sources: Au tho r’s calc ulations based on data from the Home Mortg age Disc losur e Act; Lender Processing Serv ices (LPS ) App lied Ana lytics; Rob ert Avery, Board of Govern ors of the Federa l Rese rve System;
Federa l Dep osit Insu ranc e Corp oration, Sum ma ry of Depo sits; Miss ouri Cen sus Data Center, MABLE /Ge oco rr2K : Ge ogr aph ic Corres pon den ce Eng ine with Cen sus 200 0 Geo grap hy; U.S. Bureau of Eco nom ic
Ana lysis from Hav er Ana lytic s; and U.S. Burea u of Labor Statisti cs from Hav er Analytics .
“ FICO score
Lende r share
Loan -to-in com e ratio
Loan -to-v alue ratio
FICO score
Loan denial rate
Subprim e shar e
Fixed-rate mort gage share
Jum bo shar e
Portfo lio share
Private share
Gov ernm ent shar e
1995 :Q1 a
C. Val ues for selec ted quar ters
Summary statistics, by lender type
average loan-to-value ratio for all
lenders is 83.07 percent (table 1,
panel A, third row), and it decreases
significantly over the sample period
(table 1, panel C, third row). As with
many of the other indicators, the
loan-to-value ratio suggests that
IMBs are making the riskiest loans
and local banks are making the saf
est ones. Panel B of table 1 (second
row) shows that mortgages issued
by local banks have a loan-to-value
ratio 2.44 percentage points below
the average of lenders in their mar
kets, while mortgages issued by
IMBs have a loan-to-value ratio
1.39 percentage points above the
average of lenders in their markets.
The FICO score (table 1,
panel A, fourth row) is intended to
provide a broad-based measure of
borrower quality. It includes infor
mation from the borrower’s other
loans, credit history, and other rele
vant factors. The FICO score is com
monly used to evaluate whether to
grant mortgages and other forms of
consumer credit. It ranges from 300
through 850, with a higher score
representing a safer borrower. I use
the FICO score at loan origination
as another measure of borrower
quality. The LPS data report the
FICO score starting in 1997. As
with the loan-to-income ratio, these
scores indicate that borrowers with
mortgages from IMBs are riskiest,
since they have the lowest average
FICO scores, and borrowers with
mortgages from local banks are the
safest, since they have the highest
average FICO scores (table 1,
panel A, fourth row, and panel B,
third row). In contrast to the loanto-income ratio, however, FICO
scores indicate that borrowers got
safer over time. The average FICO
score rose from 698.5 at the start
of 1997 to 715.4 at the end of 2007
(table 1, panel C, fourth row); this
trend is also noted by Bhardwaj
and Sengupta (2010) for subprime
mortgages. The differences between
9
the trends for the loan-to-income
ratio and FICO score could reflect
the difference between the risk of the
mortgage and the risk of the bor
rower prior to getting the mortgage.
The LPS data also contain an
indicator of whether a loan is con
sidered subprime (that is, loans
graded “B” or “C,” as opposed to
loans graded “A,” which are of
prime quality). As noted previously,
the LPS sample underrepresents
subprime loans. LPS data cover
about 58 percent of all loans at the
end of the sample period, in 2007,
but they only cover 33 percent of
subprime loans. Thus, the shares of
subprime lending in the data I use
should be roughly doubled to get
the share of subprime lending over
all. However, the number of sub
prime loans in the LPS data with
respect to the number of subprime
loans in mortgage-backed securities
is relatively constant over time.
Thus, while there are too few sub
prime loans in the LPS sample,
there is no reason to believe that
percentage changes in subprime loans in the LPS data
do not reflect the overall changes in subprime lending.
A mortgage is often classified as subprime because
of the low credit quality of the borrower (it also could
reflect the size of a mortgage relative to the borrower’s
ability to repay). Over the entire sample period, IMBs
issued fewer subprime mortgages than banks did (table 1,
panel A, sixth row, p. 8). Examining subprime lending
over time, I notice some interesting patterns. As illus
trated in figure 5, from 1995 through 2001 there was
little subprime lending at any type of lender. Nonlocal
banks started making a significant number of subprime
mortgages in 2002. IMBs did not start making a sig
nificant number of these loans until 2004, but when
they did, subprime mortgages went from 1 percent of
their business to 8 percent in just six months. IMBs
seemed to use subprime loans to expand, while nonlocal
banks added subprime lending at a time when their
share of lending was declining (see figure 2, p. 5).
Thus, subprime lending may have played a different
role at the two types of lenders. When the housing
market started to have troubles in 2005, IMBs were
the fastest to withdraw from the subprime mortgage
market. This is consistent with IMBs being more flex
ible than other types of lenders.
io
The measures of loan risk and borrower quality
generally indicate that borrowers with mortgages from
IMBs are riskier than those with mortgages from banks;
in addition, borrowers with mortgages from local banks
generally seem safer than those with mortgages from
nonlocal banks. There is evidence that the riskiness
of borrowers rose during the sample period, with the
largest increases during 2000-04, when home prices
were also increasing at their fastest rate (see, for ex
ample, figures 4 and 5).
I next turn to examining other mortgage features.
Mortgages come in many types, but I separate them in
two ways. First, I split fixed-rate mortgages from adjust
able-rate mortgages (ARMs).23 On average, over threequarters of all mortgages had fixed rates, but this share
moved around as mortgage rates and market conditions
changed over the sample period. More borrowers chose
fixed-rate mortgages when the yield curve was shallow
or inverted relative to when it was steep.24 The propor
tion of fixed-rate mortgages rose from 61.50 percent
in the first quarter of 1995 to 95.28 percent in the third
quarter of 1998; it edged down to 95.06 percent in the
first quarter of 2001, before falling to 56.52 percent
in the second quarter of 2005. At that point, mortgage
1Q/2011, Economic Perspectives
in its portfolio. It could sell the mort
gage to a GSE or have the mortgage
Share of subprime mortgages, by lender type
guaranteed by a government agency
(such as Ginnie Mae27) prior to sell
ing the mortgage into securitization.
Or it could sell the loan to a private
financial intermediary, often as a
prelude to securitization. Since the
selling process can take time, I use
the status of the mortgage 24 months
after origination as my measure of
whether it is held in portfolio, secu
ritized with a GSE or government
guarantee, or sold to a private firm.28
My measure may introduce a bias
because a mortgage is more likely
to end up at one of the large servicers
in the LPS data if it is securitized.
-------- Local bank
The evidence on loan sales and se
------- Nonlocal bank
curitization is likely to be indicative
------- Independent mortgage bank
of differences across the types of
lenders, but not of the true levels
Note: See the text for details on the four periods.
Sources: Author’s calculations based on data from the Home Mortgage Disclosure
of where mortgages are held. Not
Act; Lender Processing Services (LPS) Applied Analytics; Robert Avery, Board of
surprisingly, local banks hold a
Governors of the Federal Reserve System; Federal Deposit Insurance Corporation,
Summary of Deposits; and Missouri Census Data Center, MABLE/Geocorr2K:
greater percentage of their mortgages
Geographic Correspondence Engine with Census 2000 Geography.
in portfolio and, in total, sell a lower
percentage of their mortgages than
nonlocal banks and IMBs (table 1,
panel A, ninth, tenth, and eleventh rows, p. 8). The
market conditions made it more difficult for borrowers
government share, which comprises mortgages
to qualify for adjustable-rate mortgages, and the propor
securitized with a GSE or government guarantee, is
tion of fixed-rate mortgages increased to 89.19 percent
highest at nonlocal banks (table 1, panel A, eleventh
by the end of 2007 (table 1, panel C, seventh row, p. 9).
row, p. 8).
I also examine the share of so-called jumbo loans.
The loan characteristic variables are consistent
Fannie Mae and Freddie Mac were government-spon
with local banks making safer loans than other types
sored enterprises (GSEs) that purchased loans from
of lenders. This may be because they have a different
lenders prior to securitizing them.25 Fannie and Freddie
business model for mortgages, as evidenced by the fact
could only buy loans equal to or less than a given size,
that they keep a larger share of these loans in their port
known as the conforming loan limit; this limit ranged
folios. The data on borrower qualify and loan character
from $203,150 at the start of the sample, in 1995, to
istics suggest systematic differences across lender types.
$417,000 at the end of the sample, in 2007.26 Loans that
otherwise are of prime qualify but are larger than the con
Impact of lender types on local market
forming loan limit are known as jumbo loans. For much,
lending standards
if not all, of the sample period, jumbo loans were more
In this section, I extend the examination of whether
difficult to securitize than conforming loans. Thus, they
mortgage lending and mortgage terms in a local mar
were more likely to be kept in a lender’s portfolio. This
ket are related to the types of lenders in that market.
may make it unsurprising that local banks made the
In the previous section, I showed that mortgage lend
largest share of jumbo mortgages (table 1, panel A,
ing standards are correlated with the market shares of
eighth row, p. 8).
different types of lenders. But the simple statistics do
As indicated in the prior paragraph, lenders were
not allow us to determine whether the presence of
able to sell certain mortgages to Fannie Mae and
one type of lender affects the mortgages offered by
Freddie Mac. In general, a lender had three options
other types of lenders. Here, I use a regression model
when it issued a mortgage. It could hold the mortgage
to tease this out.
FIGURE 5
Federal Reserve Bank of Chicago
11
The baseline model allows lending standards to
be a function of lender types and market conditions:
1)
Lending standards.ct =/{Lendershares.ct
Lending market conditionsct Economic
conditionsc,t, l7’
where z is the type of lender (local bank, nonlocal
bank, or IMB), c refers to the local market (county),
and t is the time period. The right-hand side variables
are all lagged one quarter to mitigate potential endo
geneity problems.29
The characteristics I examine are those that focus
on lending standards. The loan-to-income ratio and
the loan-to-value ratio are direct measures of loan
risk, while the FICO score and the share of subprime
loans are measures of borrower quality (of course, a
high-quality borrower with a high FICO score can
nonetheless take a risky loan—for example, one with
a very high loan-to-income ratio). Classifying the loan
denial rate along these lines is more difficult. Loans
can be denied either because a borrower has a weak
profile or because the loan is too risky given the qual
ity of the borrower. Thus, it mixes loan risk and borrower
risk. Each of these characteristics can be affected by
competitive conditions in a market, which include the
different incentives of each type of lender.
I use each lending standard as both a dependent
variable and a control because each can pick up aspects
of market conditions other than differences across
lenders. A high average loan-to-income ratio can re
flect borrowers needing to commit a larger share of
income in order to purchase a home in markets where
homes are relatively expensive. Similarly, expensive
homes may reduce the percentage down payment that
borrowers can make, leading to a higher loan-to-value
ratio. Additionally, in the recent crisis, some borrowers
with loan-to-value ratios above 100 percent have walked
away from their mortgages because they have negative
equity in their homes. The risk of this happening is
obviously higher when a mortgage has a larger initial
loan-to-value ratio. More lender competition can reduce
average FICO scores or lead to fewer loans being de
nied.30 Similar to the loan-to-income ratio, the share
of subprime loans in a market may be correlated with
home prices in the market. Of course, it can also be
affected by competition among lenders and changes
in securitization markets.
Some aspects of loan quality that have a weaker
correlation with loan risk are included as controls but
not as dependent variables. The share of loans kept in
portfolio is likely to be related to the types of lenders
in a market. There may be a weak correlation with
12
risk because it is more difficult to securitize unusual
loans. The proportion of fixed-rate mortgages may
reflect borrower strength, especially in later years when
borrowers often qualified for mortgages based on their
ability to meet the initial loan payments. The ability
of borrowers of a given income and risk to qualify
for larger adjustable-rate mortgages than fixed-rate
mortgages means that, all else being equal, fixed-rate
mortgages were safer to fund.
A number of the lending market standard variables
are affected by the ability of potential borrowers to
purchase a home. I control for prices in two ways. First,
I include the percentage change in home prices over
the past quarter in the local market (so the change in
period t - 1 is the percentage difference from period
t - 2 to period t - 1). I measure prices using the FHFA
HPI. There is an extensive debate in the housing liter
ature about what the best price index is (see Rosen,
2008; and Case and Shiller, 2003). I choose the FHFA
HPI because it is available for a wider number of mar
kets than other constant-quality indexes, such as the
Standard and Poor’s/Case-Shiller Home Price Index.
The second control I use is the price-to-rent ratio in
the local market. I measure rents using the owners’
equivalent rent component of the Consumer Price
Index (CPI-OER), which is put out by the U.S. Bureau
of Labor Statistics. The price-to-rent ratio is, thus, the
ratio of the FHFA HPI to the CPI-OER. A high value
indicates that owning a home is expensive relative to
renting. For both controls, I use the data for the MSA
that a market is in if available. Otherwise, statewide
data are used.
I also add additional controls for local economic
conditions. These include measures of the unemploy
ment rate and income per capita.31 For both variables,
I use the mean value for the MSA a county is in if
that is available. Otherwise, I use the mean value for
the state. To pick up any systematic local differences
not captured by the other controls, I include countylevel dummies in the main regression.
There were secular trends in many of the lending
market standards; for example, the rise of securitiza
tion and the increased use of the “originate-to-distribute” model for mortgages during the run-up in home
purchases (see figure 1, p. 5) affected the mortgages
lenders issued (see, for instance, Keys et al., 2010).
Such trends may have given lenders an incentive
to issue high loan-to-income, high loan-to-value, or
low FICO-score mortgages. To control for the com
mon effects of the rise and fall of securitization,
I include time dummies in the regressions. The time
dummies also pick up other changes in lending tech
nology, economic conditions, and interest rates that
1Q/2011, Economic Perspectives
Finally, since one objective is to
examine how the distribution of lender
Effect of mortgage and market characteristics
types affects loan characteristics, I ex
on loan-to-income ratio in local markets,
clude some small markets. To be included,
by lender type
a county must average 50 loans per quarter,
Loan-to-income ratio at:
with an average of at least live by each
Independent
type of lender (local banks, nonlocal banks,
Local
Nonlocal
mortgage
and IMBs). The final data set includes
banks
banks
banks
observations for all county-quarters with
Local bank share
0.223*
0.073
-0.250***
mortgage market and local economic data.
(0.057)
(0.382)
(0.002)
There are 31,010 observations, covering
Nonlocal bank share
-0.259"
0.175"
-0.320***
800 counties during 52 quarters.32 This is
(0.012)
(0.014)
(0.000)
an
unbalanced panel, since newly created
Loan-to-value ratio
-0.077
0.138
0.212
counties are added when they appear in
(0.685)
(0.527)
(0.284)
the data.
FICO score
-0.001"
-0.000
-0.000
(0.048)
(0.146)
(0.415)
One issue with using aggregate lend
ing market standards is that it is not pos
Loan denial rate
-0.002
-0.047
0.097
(0.986)
(0.377)
(0.201)
sible to determine whether the resultant
Subprime share
0.292
0.003
-0.319"
correlations reflect the effect of competition
(0.103)
(0.980)
(0.029)
among lenders as opposed to just a change
Portfolio share
-0.026
-0.058
0.005
in the mix of lenders. To focus on the re
(0.775)
(0.531)
(0.958)
lationship between the mortgage shares of
Fixed-rate mortgage share -0.027
-0.029
-0.044
different lender types and the characteristics
(0.657)
(0.542)
(0.609)
of
mortgages, I separately consider mort
Unemployment rate
0.131
-0.070
0.653"
gages by each type of lender in a market.
(0.823)
(0.821)
(0.047)
That is, for each lending characteristic,
Income per capita
-0.000*
0.000*
-0.000"
(0.096)
(0.078)
(0.017)
I run separate regressions for the average
characteristics of local banks, nonlocal
Change in home price
-0.353
-0.580
0.050
(0.166)
(0.312)
(0.827)
banks, and IMBs.
Price-to-rent ratio
0.892***
0.831***
0.814***
Table 2 presents the coefficient esti
(0.000)
(0.000)
(0.000)
mates for regressions of equation 1, using
Adjusted R-squared
0.513
0.391
0.373
the loan-to-income ratio for the mortgages
p value for test of local bank
that each type of bank has made as the
share = nonlocal bank share 0.000
0.021
0.389
dependent variable. The first column re
*p<0.10
ports the results for local banks. The posi
**p < 0.05
tive sign on the coefficient for the local
***p<0.01
bank share of the market (first row) im
Notes: FICO score indicates the Fair Isaac Corporation credit score. Full definitions
for the variables are in the text. The regression in the first column has the loan-toplies that as the proportion of mortgages
income ratio at local banks as a dependent variable. The regression in the second
in a market issued by local banks increas
column has the loan-to-income ratio at nonlocal banks as a dependent variable.
The regression in the third column has the loan-to-income ratio at independent
es, the average loan-to-income ratio on
mortgage banks as a dependent variable. Results in parentheses directly below
the regression coefficients are p values (of statistical difference from zero). The
all mortgages issued by local banks in that
test values reported in the final row are p values for a test that the local bank
market
increases. Shifting the mortgage
share coefficient is equal to the nonlocal bank share coefficient. Each regression
has 31,010 observations.
share from IMBs (the omitted variable)
Sources: Author’s calculations based on data from the Home Mortgage Disclosure
to local banks is associated with an increase
Act; Lender Processing Services (LPS) Applied Analytics; Robert Avery, Board of
Governors of the Federal Reserve System; Federal Deposit Insurance Corporation,
in the loan-to-income ratio, with a one
Summary of Deposits; Missouri Census Data Center, MABLE/Geocorr2K:
Geographic Correspondence Engine with Census 2000 Geography; U.S. Bureau of
standard deviation increase in the local
Economic Analysis from Haver Analytics; U.S. Bureau of Labor Statistics from Haver
bank share (3.12 percent, as given in table 1,
Analytics; and Federal Housing Finance Agency, seasonally adjusted purchase-only
House Price Index, from Haver Analytics.
panel A, first row, p. 8) implying a 0.70
percent increase (3.12 x 0.223), or about
2.4 percent of its mean (0.70/28.46). The
coefficient on the nonlocal bank share is negative (sec
are common across markets. Including time dummies
ond row of table 2). This means that shifting loan share
helps me focus on how the loan shares of different
from IMBs to nonlocal banks reduces the average
lender types affect lending market standards.
TABLE 2
Federal Reserve Bank of Chicago
13
loan-to-income ratio at local banks in the market. Also,
the coefficients on the local bank share and the nonlocal
bank share (first and second rows of table 2) are signif
icantly different from one another (as shown in the fi
nal row of the table, which gives the p value for a test
that the two coefficients are equal). So, a movement
in lending from nonlocal banks to local banks is asso
ciated with significant increases in the loan-to-income
ratio at local banks.
The results for the loan-to-income ratio at nonlocal
banks and IMBs are presented in the second and third
columns of table 2. One common element in all three
regressions is that when the loan market share of a
particular type of lender is increasing, the average loanto-income ratio of mortgages from that type of lender
increases. This is indicated by the positive coefficients
on local bank share in the local bank regression (first
row, first column) and on nonlocal bank share in the
nonlocal bank regression (second row, second column).
It is also indicated by the negative coefficients on both
bank shares in the IMB regression (first and second rows,
third column); both local and nonlocal bank shares
decreasing means that the IMB share is increasing.
When one type of lender increases its market share
in period t - 1, mortgages from that type of lender are
riskier, all else being equal, in period t. As described
previously, the loan-to-income ratio for mortgages
issued by local banks changes when there is a shift
in market share between nonlocal banks and IMBs
(second row, first column). This ratio, however, does
not change for mortgages issued by nonlocal banks
when there is a shift in market share between local
banks and IMBs (first row, second column). Similarly,
mortgages issued by IMBs do not change their loanto-income ratio when market share shifts between lo
cal and nonlocal banks (final row, third column).
I now briefly discuss the coefficients on the other
control variables in table 2. These are representative
of the coefficients on later regressions. There is gen
erally only a weak correlation among the measures
of borrower quality. For example, in table 2, the coef
ficients on FICO score (fourth row) and the subprime
share (sixth row) are each significant in only one re
gression, while the coefficients on the loan-to-value
ratio (third row) and the loan denial rate (fifth row)
are not significant for any of the regressions.
Changes in some of the macroeconomic factors
featured in table 2 can affect the loan-to-income ratio
at the different lender types. Higher income (tenth row)
is associated with an increase in the loan-to-income
ratio in the mortgages made by nonlocal banks, but a
reduction in the loan-to-income ratio in the mortgages
made by local banks and IMBs. This may indicate that
14
changes in local income are associated with shifts among
lender types. When the price-to-rent ratio (twelfth row)
increases, buying a home is relatively more expensive
than renting one. This makes it likely that when people
do buy a home, they are not able to afford a large down
payment, and thus they have a large loan-to-income ratio.
Table 3 presents the coefficients on the lender
share variables for regressions of equation 1, using
the averages of the loan characteristics by lender type
as the dependent variables. This repeats the regressions
in table 2 and also includes regressions where the de
pendent variables are the loan-to-value ratio, FICO
score, loan denial rate, and subprime mortgage share.
The other controls, although not shown, are the same
as those for the regressions in table 2.
Two patterns are apparent from table 3. First,
changes in lender shares have a different impact on
loan risk characteristics than on borrower quality
characteristics (here, the loan denial rate looks more
similar to a loan risk characteristic than a borrower
quality characteristic). While changes in lender shares
are associated with riskier mortgages as measured by
the loan risk indicators, such changes are associated
with less risky mortgages as measured by borrower
quality indicators. For example, an increase in the local
bank share is associated with smaller loan-to-income
and loan-to-value ratios and a larger loan denial rate
at IMBs (third column), all indicating less risky mort
gages. However, this increase is also associated with
lower FICO scores and more subprime lending, which
indicate lower-quality borrowers. One possible expla
nation is that high-quality borrowers were taking out
risky loans; that is, borrowers with higher FICO scores
took out loans that were risky enough to be classified
subprime. Consistent with this interpretation, others have
documented that FICO scores of subprime loans have
increased since 2000 (Demyanyk and Van Hemert,
2009; and Bhardwaj and Sengupta, 2010). However,
an analysis of why direct measures of loan risk seem to
move in the opposite direction as measures of borrower
quality is beyond the scope of this article.
A second pattern in table 3 is that, as a particular
type of lender increases market share, the loans made
by that type of lender tend to get riskier. As noted pre
viously, the loan-to-income ratio for mortgages made
by a lender type is larger as the own-type lender share
increases.33 The loan-to-value ratio increases and the
share of loans denied decreases in these circumstances.
The picture for subprime shares is mixed, with local
banks (first column) having a larger share of subprime
lending when local bank share increases, but nonlocal
banks and IMBs (second and third columns) having the
opposite reaction to own-type lender share increases
1Q/2011, Economic Perspectives
TABLE 3
Effect of mortgage and market characteristics on loan risk and borrower quality
in local markets, by lender type
Independent
Dependent variable
Loan-to-income ratio
Independent variable
Local
Nonlocal
Test: Local = Nonlocal
Loan-to-value ratio
Local
Nonlocal
Local
Nonlocal
Test: Local = Nonlocal
Loan denial rate
Local
Nonlocal
Test: Local = Nonlocal
Subprime share
Nonlocal banks
mortgage banks
0.223*
-0.259"
0.073
0.175"
-0.250***
-0.320***
0.000
Test: Local = Nonlocal
FICO score
Local banks
Local
Nonlocal
Test: Local = Nonlocal
0.021
0.007
0.066***
0.389
0.000
0.000
-0.075***
-0.052***
0.035
104.930"
10.050
11.686
15.958
-144.131***
-81.212***
0.013
0.726
-0.076***
-0.016
-0.071***
-0.176***
0.000
0.000
0.026"
-0.017*
0.000
-0.016"
0.000
0.018
0.078*
-0.077"
0.023
0.255***
0.308***
0.021
0.026***
0.023***
0.522
*p <0.10
**p < 0.05
***p < 0.01
Notes: FICO score indicates the Fair Isaac Corporation credit score. Full definitions for the variables are in the text. Coefficients on lender share
variables are reported. Local is the coefficient on the local bank share, and nonlocal is the coefficient on the nonlocal bank share (independent
mortgage bank share is the omitted variable). The regressions on which these are based include all the control variables for the regressions
reported in table 2. The dependent variables for these regressions in the local banks column are the local bank average for the variable given in
the leftmost column. Other dependent variables are similarly defined. The test values reported are p values for a test that the local bank share
coefficient is equal to the nonlocal bank share coefficient. All regressions except those with FICO score as the dependent variable have 31,010
observations. The regressions with FICO score as the dependent variable have 26,445 observations.
Sources: Author’s calculations based on data from the Home Mortgage Disclosure Act; Lender Processing Services (LPS) Applied Analytics;
Robert Avery, Board of Governors of the Federal Reserve System; Federal Deposit Insurance Corporation, Summary of Deposits; Missouri
Census Data Center, MABLE/Geocorr2K: Geographic Correspondence Engine with Census 2000 Geography; U.S. Bureau of Economic Analysis
from Haver Analytics; U.S. Bureau of Labor Statistics from Haver Analytics; and Federal Housing Finance Agency, seasonally adjusted purchaseonly House Price Index, from Haver Analytics.
(see note 33). Consistent with the differences between
own-type share changes and other-type share changes,
there is generally a statistically significant difference
between the coefficients on the local bank share and
the nonlocal bank share (p values for these tests are
reported in the table).
Lending standards at local banks seem to shift more
after there are changes in nonlocal bank share, com
pared with the lending standards at nonlocal banks fol
lowing changes in local bank share. To see this, compare
the coefficients on nonlocal bank share in the first
column with the coefficients on local bank share in
second column of table 3. This shows that a shift in
mortgage shares from IMBs to nonlocal banks is
associated with a decrease in the risk of mortgages
issued by local banks, while a shift from IMBs to local
banks has little impact on the risk of mortgages issued by
nonlocal banks. For instance, when the nonlocal bank
share increases, the loan-to-income and loan-to-value
ratios for mortgages issued by local banks decrease,
Federal Reserve Bank of Chicago
indicating safer loans (table 3, second and fifth rows,
first column). However, an increase in the local bank
share has no significant impact on these ratios for
mortgages made by nonlocal banks (table 3, first
and fourth rows, second column).
It is instructive to compare the results in table 3
with those in panel A of table 1 (p. 8). As shown in
panel A of table 1, mortgages issued by local banks
have the lowest loan-to-income and loan-to-value
ratios. Yet, as the coefficients on local bank share
in the first column of table 3 show, when local bank
lender share increases in a market, loans issued by
local banks tend to have higher risk (that is, higher
loan-to-income ratios, higher loan-to-value ratios,
a greater likelihood to be subprime, and lower loan
denial rates). In addition, as market share shifts from
IMBs to nonlocal banks, mortgages issued by nonlocal
banks generally increase in risk, as indicated by the
coefficients on nonlocal bank share in the second
column of table 3. Specifically, the loan-to-income
15
and loan-to-value ratios increase and the
loan denial rate decreases, consistent with
riskier lending practices (although the share
of subprime loans decreases, pointing in
the other direction). One interpretation
consistent with this is that lenders compete
more with lenders of the same type than
lenders of other types, and competition
manifests itself in allowing borrowers to
take larger loans relative to both borrower
income and home values. Of course, this
does not necessarily mean that lenders are
providing mortgages to riskier borrowers.
Borrowers with mortgages from local
banks have the highest FICO scores, but
competition among local banks does not
seem to lower the average FICO score of
borrowers who get their mortgages from
local banks.
Since the proportion of local bank
lending fell during most of the sample pe
riod, until the housing crisis started in late
2005 (recall figure 2, p. 5), we can think
about how this might have changed lend
ing standards. As local banks made fewer
loans in a market, loan risk decreased at
local banks and increased at IMBs. To the
extent that lender share by local banks
was lost to nonlocal banks and IMBs, the
net effect on loan risk at nonlocal banks
was small. It is important to remember
that there are time dummies in these re
gressions, so any changes are above and
beyond secular trends across lender types.
TABLE 4
Summary statistics for counties in large and
small metropolitan statistical areas (MSAs)
Top 50
(large) MSAs
Mean
Standard
deviation
Non-top-50
(small) MSAs
Mean
Standard
deviation
Local bank share
22.3
16.8
26.4
22.5
Nonlocal bank share
44.9
17.9
45.2
22.1
Independent mortgage
bank share
32.7
13.9
28.5
17.2
Loan-to-income ratio
2.36
0.40
2.13
0.50
Loan-to-value ratio
83.1
5.5
84.6
5.0
705.17
19.75
702.61
20.85
17.8
8.5
22.3
11.1
FICO score
Loan denial rate
Subprime share
2.6
3.7
2.8
4.6
78.9
16.6
84.7
13.4
Jumbo share
7.3
12.0
2.5
6.2
Portfolio share
7.7
6.6
6.7
7.1
Private share
22.0
12.3
18.9
12.9
Government share
70.4
14.8
74.4
14.3
Fixed-rate mortgage share
Unemployment rate
Income per capita
Share of market
4.6
1.2
4.9
2.0
33,932
7,146
27,660
6,049
59.67
40.33
Notes: All values are in percent except those for loan-to-income ratio; FICO score,
which indicates the Fair Isaac Corporation credit score; and income per capita, which
is in dollars. Full definitions for the variables are in the text. A county is considered
a large-MSA county if it is in one of the 50 largest metropolitan divisions/MSAs,
according to the 2000 U.S. Census. Otherwise, it is considered a small-MSA county.
See the text for further details. Certain shares may not total because of rounding.
Sources: Author’s calculations based on data from the Home Mortgage Disclosure
Act; Lender Processing Services (LPS) Applied Analytics; Robert Avery, Board of
Governors of the Federal Reserve System; Federal Deposit Insurance Corporation,
Summary of Deposits; Missouri Census Data Center, MABLE/Geocorr2K: Geographic
Correspondence Engine with Census 2000 Geography; U.S. Bureau of Economic
Analysis from Haver Analytics; and U.S. Bureau of Labor Statistics from Haver Analytics.
Lending standards and
market size
The markets in the sample range from small
counties with populations of less than 50,000 all the
way up to the New York City area with over 10 mil
lion residents. To see whether lenders compete the
same way in the large metropolitan areas as elsewhere,
I divide the sample of counties (markets) into two
categories. I place counties in MSAs. For very large
MSAs, I further divide them into metropolitan divisions.
Metropolitan divisions are groups of closely tied con
tiguous counties that serve as distinct employment
districts. They are part of MSAs with populations of
at least 2.5 million. I define a county as a large-MSA
county if it is in one of the top 50 metropolitan divi
sions/MSAs; otherwise, I define a county as a smallMSA county. The MSAs are ranked by population
according to the 2000 U.S. Census. The largest metro
16
area is the New York-Wayne-White Plains, NY-NJ
metropolitan division and the fiftieth largest is the
Memphis, TN-MS-AR MSA. Table 4 presents a com
parison of large-MSA and small-MSA counties. There
are some significant differences between mortgage
market conditions across large-MSA and small-MSA
counties. However, it is not clear that one type of
county has riskier conditions than the other type.
To see whether the differences between markets
in large and small MSAs affect competition among
lenders, I split the lender share variables by whether a
market is a large-MSA or small-MSA county. Table 5
presents results for regressions including these variables.
The regressions include the same nonlender control
variables as the regressions in tables 2 and 3, but only
the coefficients on the lender share variables are reported.
1Q/2011, Economic Perspectives
Federa l Res erv e Bank of Chicag o
TABLE 5
Effect of mortgage and market characteristics on loan risk and borrower quality in local markets,
by lender type and metropolitan statistical area (MSA) size
Local banks
Dependent variable
Loan-to-income ratio
Independent variable
Local
Nonlocal
Test: Local = Nonlocal
Loan-to-value ratio
Local
Nonlocal
Test: Local = Nonlocal
FICO score
Local
Nonlocal
Test: Local = Nonlocal
Loan denial rate
Local
Nonlocal
Test: Local = Nonlocal
Subprime share
Test: Local = Nonlocal
Small
Test; L = S
0.056
-0.572***
0.315**
-0.097
0.151
0.011
0.000
0.000
0.009
-0.182***
0.115**
-0.023
0.003
0.000
168.303***
-37.756
86.511*
34.080
0.202
0.240
0.419
-0.084***
-0.029**
0.000
0.241
0.088
0.045***
-0.003
0.000
0.003
0.048
0.003
-0.060***
0.009
0.000
Local
Nonlocal
-0.018
-0.044**
0.018
Independent mortgage banks
Nonlocal banks
Large
0.109
0.016
Large
Small
Test: L = S
Large
Small
Test; L = S
-0.285***
-0.296***
-0.239**
-0.331***
0.756
0.780
0.880
0.364
-0.042**
-0.063***
-0.086***
-0.045***
0.005
0.084
0.527
-185.701***
-82.374**
0.002
0.954
0.001
-0.112
0.085
0.126
0.321**
0.000
0.449
0.020
0.012
-0.022
0.013
0.110***
0.000
0.955
0.000
0.023
0.142
0.871
0.080
16.403
-7.369
12.522
28.427
0.269
0.268
-0.097***
-0.110***
0.453
0.026
0.707
-0.067**
-0.209***
0.000
0.004
0.008
-0.003
-0.027***
0.664
0.010
0.786
0.003
-32.116
-84.943**
0.148
0.003
0.133***
0.284***
0.000
0.304***
0.324***
0.005
0.023***
0.034***
0.023***
0.087
0.040
0.000
0.362
0.477
0.022
0.985
*p < 0.10
**p < 0.05
***p< 0.01
Notes: FICO score indicates the Fair Isaac Corporation credit score. Full definitions for the variables are in the text. Local is the coefficient on the local bank share, and nonlocal is the coefficient on the nonlocal
bank share (independent mortgage bank share is the omitted variable). The regressions on which these are based include all the control variables for the regressions reported in table 2. The dependent variables
for these regressions in the local banks columns are the local bank averages for the variable given in the leftmost column. Other dependent variables are similarly defined. The values reported for “Test: Local =
NonlocaT are p values for a test that the local bank share coefficient is equal to the nonlocal bank share coefficient. The values reported for “Test: L = S” are p values for a test that the bank share coefficient for
large-MSA counties is equal to the bank share coefficient for small-MSA counties (see the text and table 4 for definitions of the large-MSA and small-MSA counties).
Sources: Author’s calculations based on data from the Home Mortgage Disclosure Act; Lender Processing Services (LPS) Applied Analytics; Robert Avery, Board of Governors of the Federal Reserve System;
Federal Deposit Insurance Corporation, Summary of Deposits; Missouri Census Data Center, MABLE/Geocorr2K: Geographic Correspondence Engine with Census 2000 Geography; U.S. Bureau of Economic
Analysis from Haver Analytics; U.S. Bureau of Labor Statistics from Haver Analytics; and Federal Housing Finance Agency, seasonally adjusted purchase-only House Price Index, from Haver Analytics.
The pattern of responses to mortgage share changes
in both groups of markets is similar to that of the full
sample—with one exception. In large-MSA markets,
lenders, especially local and nonlocal banks, seem to
react less to changes in market share of lenders of the
same type. For example, a change in local bank share
is not associated with a significant change in the loanto-income ratio or the loan-to-value ratio of mortgages
issued by local banks (the coefficients 0.056 and 0.009,
as given in the first column of table 5, are not significantly
different from zero); also, a change in local bank share is
not associated with a significant change in the share
of subprime lending at the local banks (the coefficient
-0.018 in the first column of table 5 is not significantly
different from zero). All these coefficients are statisti
cally significant in the similar regression for the sample
as a whole (table 3, p. 15). This suggests that compe
tition may be more complex in counties that are part
of large MSAs.
Table 5 also presents tests of differences in the re
gression coefficients across large-MSA and small-MSA
markets. The p values reported in the columns labeled
“Test: L = S” are for tests of the differences between the
coefficients in the large-MSA regressions and those in
the small-MSA regressions. These results show little dif
ference in how banks react to changes in local bank mar
ket share based on MSA size. However, in large-MSA
markets relative to small-MSA markets, changes in
nonlocal bank lender share are generally associated with
larger changes in local bank mortgage characteristics but
smaller changes in nonlocal bank mortgage character
istics. Again, this is consistent with differences in how
lenders compete across MSAs that differ in size.
The results presented in tables 3 and 5 show that
the distribution of lender types affects lending standards
and loan characteristics. During the housing boom,
the share of lending by local banks decreased, since
both nonlocal banks and IMBs increased their market
share. All else being equal, this means that loan risk
and borrower quality fell for mortgages made by local
banks, even more so in small-MSA markets than in
large-MSA markets. To the extent that lending migrated
to nonlocal banks from IMBs, loan risk and borrower
quality increased for mortgages made by nonlocal banks,
again more so in small-MSA markets than in largeMSA markets. As the share of loans made by IMBs
increased, loan risk and borrower quality increased
at IMBs, in both large-MSA and small-MSA markets.
18
Conclusion
I examine mortgage lending during the period
1995-2007. This was a period of extensive change in
the mortgage market. There was a boom and bust in
home purchases and home prices. What caused the
boom and bust is a big question that is still being de
bated. One possible contributing factor is the shift in
the mortgage delivery process. During the housing boom,
fewer and fewer borrowers got their mortgages from
local banks; both nonlocal banks and IMBs gained mar
ket share. This could have affected mortgage markets
because each type of lender approaches mortgage
lending differently. Local banks have a more intensive
retail focus and are most likely to keep loans in port
folio. Banks that make loans outside their local mar
kets (nonlocal banks) are likely to use the wholesale
lending channel for these loans, but being banks, they
sometimes will keep loans in portfolio. In contrast,
IMBs are wholesale lenders that sell essentially all
the loans they originate.
The changes in market shares of lender types could
be important because the characteristics of mortgages
are a function of the lender type. Local banks tend to
make loans that appear ex ante safer—for example, they
have lower loan-to-income and loan-to-value ratios.
Thus, the market shift away from mortgages issued by
local banks could lead to riskier mortgages being made.
The shift in lenders can also have an indirect
effect. In part, loan characteristics for mortgages
made by a particular type of lender may depend not
only on that type of lender’s cost-benefit trade-off,
but also on the competitors it faces. I show that an
increase in the mortgage market share of a particular
type of lender is associated with other lenders of the
same type increasing the average loan risk of their
mortgages; at the same time, this increase in the
mortgage market share of a particular type of lender
is associated with an increase in the average quality
of their borrowers. This impact is larger in counties
that are in small MSAs.
My analysis suggests that the efforts to get (private)
mortgage securitization markets going again might
affect the types of mortgages that are issued because
of their effects on lender composition. The securitiza
tion market facilitates the wholesale lending channel,
and is likely to increase the share of loans made by
nonlocal banks and IMBs. These loans tend to be riskier
on average than loans made by local banks. In addition,
the indirect effect of changing the market structure
may be to increase loan risk even further at nonlocal
banks and IMBs, although not at local banks. Hence,
both the direct and indirect effects may add to aggre
gate loan risk.
1Q/2011, Economic Perspectives
NOTES
’For example, see Steverman and Bogoslaw (2008).
2Subprime lending is the issuing of loans to borrowers with poor
or no credit histories; mortgage securitization is the packaging and
sale of bonds that have mortgages as the underlying collateral. In
addition, see U.S. Congress, Joint Economic Committee (2007)—
a report on the housing crisis that centers around subprime lending.
3Inside Mortgage Finance Publications (2008).
4Previous work has examined how the structure of the mortgage
industry has affected discrimination in lending (see Apgar,
Bendimerad, and Essene, 2007).
5This is similar to the approach in the literature examining how
the size and organizational structure of competitors in banking
markets can affect deposit rates and small market lending (Rosen,
2007a; Berger, Rosen, and Udell, 2007; and Park and Pennacchi, 2009).
6The values cited here are from my calculations based on data from
the Home Mortgage Disclosure Act; Robert Avery, Board of
Governors of the Federal Reserve System; and Federal Deposit
Insurance Corporation, Summary of Deposits.
7One would expect that brokers would lead borrowers to the lender
offering the best deal. However, there are allegations that some brokers
steered borrowers toward loans that maximized the brokers’ com
missions rather than minimized the borrowers’ costs (see, for example,
the comments of Senator Christopher J. Dodd, D-CT, in 2007 at
http://dodd.senate.gov/index.php?q=node/4167).
8See Wholesale Access Mortgage Research and Consulting Inc. (2005).
9There is an intermediate case, where a small lender originates a
loan and then quickly sells it to a large wholesale lender under
prearranged terms. See Apgar, Bendimerad, and Essene (2007)
for a more detailed discussion of the different origination channels.
10This is derived from the Home Mortgage Disclosure Act (HMDA)
data described later in the article.
11 An important feature of the FICO score is that it is intended to
measure a borrower’s creditworthiness prior to taking out a mort
gage. FICO scores range between 300 and 850. Typically, a FICO
score above 800 is considered very good, while a score below 620
is considered poor. As reported on the Fair Isaac Corporation web
site (www.myfico.com), in June 2009 borrowers with FICO scores
above 760 were able to take out 30-year fixed-rate mortgages, or
FRMs (see note 20), at interest rates that were 160 basis points
lower, on average, than those available for borrowers with scores
in the 620-639 range.
12Later in the article, I explain exactly how I divide the sample into
large-MSA counties and small-MSA counties.
this means that the history of lender activity may be more important
for refinancings than for purchase loans. Also excluded are home
equity lines, which are revolving lines of credit with a home serv
ing as collateral. Since these loans are not generally completely
drawn at initiation, their pricing and characteristics may vary from
those of basic mortgages.
15The different types of depository institutions reflect differences in
their charters and regulators, as well as historical differences in the
types of loans they issue. A commercial bank’s primary federal reg
ulator is the Office of the Comptroller of the Currency, the Federal
Reserve, or the Federal Deposit Insurance Corporation. Thrift
banks are regulated by the Office of Thrift Supervision; and credit
unions are regulated by the National Credit Union Administration.
16The classification is based on a data set provided by Robert Avery,
Board of Governors of the Federal Reserve System.
17The HPI is an index based on repeat sales information. It comes
from the FHFA, which was established in 2008 by the Federal Housing
Finance Regulatory Reform Act of 2008, a part of the Housing and
Economic Recovery Act of 2008. The FHFA was formed by a merger
of the Office of Federal Housing Enterprise Oversight (OFHEO),
the Federal Housing Finance Board, and the U.S. Department of
Housing and Urban Development’s government-sponsored enter
prise mission team (see www.fhfa.gov for additional details). The
HPI was formerly published by OFHEO.
18I have no information on branch locations for credit unions, so
I assume all mortgages made by credit unions are in markets where
they have branches (that is, I assume all mortgages issued by credit
unions are local bank mortgages).
19A small number of loan applications that are approved but not
taken are dropped from this calculation.
20A fixed-rate mortgage is one whose interest rate is fixed from its
origin for its entire term.
21The LPS data include the ratio of the initial mortgage payment to
the borrower’s monthly income from 2005 on. The cross-sectional
pattern of the data is similar to that for the loan-to-income ratio in
the HMDA data.
22I drop all observations where the loan-to-value ratio is above
250 percent, as these likely represent data errors.
23Unlike an FRM, whose interest rate is fixed from its origin for its
entire term, an ARM’s interest rate can adjust periodically based
on terms set in the mortgage contract. When an ARM resets after
an initial defined period (which may be as short as one year or as
long as seven), the interest rate and, consequently, the monthly
mortgage payment may change substantially.
13For details, see Federal Financial Institutions Examination Council
(2008). In general, very small lenders are exempt from filing, as
are lenders that do not make loans in metropolitan statistical areas.
24A yield curve shows the relationship between yields and maturity
dates for a set of similar bonds, usually Treasuries, at a given point
in time. A steep yield curve means that ARMs tend to have much
lower initial interest rates than do FRMs; the interest differential is
small when the yield curve is relatively flat.
14The major excluded group is loans to refinance existing mortgages.
The share of loans that are for refinancing varies over time, influ
enced in large part by the pattern of mortgage interest rates. I exclude
these loans for two main reasons. First, the exclusion makes it easier
to determine the role played by the lender, since I do not have to
control for changes in the mix of loans. Second, borrowers’ current
lenders may have an advantage in capturing refinancing loans, and
25The full official name for Fannie Mae is the Federal National
Mortgage Association. The full official name for Freddie Mac
is the Federal Home Loan Mortgage Corporation. The two
government-sponsored enterprises were put into conservatorship
in 2008.
Federal Reserve Bank of Chicago
19
26This is the limit for a single-family home, which was set by the
OFHEO and is now set by the FHFA. There were higher limits for
multifamily homes.
27The full official name of Ginnie Mae is the Government National
Mortgage Association.
28For mortgages that leave the data prior to 24 months (which often
reflects repayment or default), I use the status in the last month the
mortgage is in the data to measure its disposition.
29Including additional lags does not qualitatively change the results.
31Income per capita is only available at an annual frequency.
I linearly interpolate across quarters. The data come from the
U.S. Department of Commerce and are based on population esti
mates by the U.S. Census Bureau. The unemployment data are
from the U.S. Bureau of Labor Statistics.
32Our restrictions on the number of loans eliminate 115 smaller
counties from the sample.
33To find what happens when the own-type lender share for IMBs
increases, one would have to take the negative of the reaction to
an increase in the lender shares of local and nonlocal banks.
30FICO scores are only available from 1997 onward. For earlier
years, the FICO score variable is set to zero when it is used as a
control. In these years, the average FICO score for the nation is
captured by time dummies.
20
1Q/2011, Economic Perspectives
REFERENCES
Apgar, William, Amal Bendimerad, and Ren S.
Essene, 2007, “Mortgage market channels and fair
lending: An analysis of HMDA data,” Harvard
University, Joint Center for Housing Studies, working
paper, April 25, available at www.jchs.harvard.edu/
publications/finance/mm07-2_mortgage_market_
channels.pdf.
Berger, Allen N., Richard J. Rosen, and Gregory
F. Udell, 2007, “Does market size structure affect
competition? The case of small business lending,”
Journal ofBanking and Finance, Vol. 31, No. 1,
pp. 11-33.
Bhardwaj, Geetesh, and Rajdeep Sengupta, 2010,
“Where’s the smoking gun? A study of underwriting
standards for U.S. subprime mortgages,” Federal
Reserve Bank of St. Louis, working paper,
No. 2008-036D, revised October 2010.
Keys, Benjamin J., Tanmoy Mukherjee,
Amit Seru, and Vikrant Vig, 2010, “Did securitiza
tion lead to lax screening? Evidence from subprime
loans,” Quarterly Journal of Economics, Vol. 125,
No. 1, February, pp. 307-362.
Mian, Atif, and Amir Sufi, 2009, “The consequences
of mortgage credit expansion: Evidence from the U.S.
mortgage default crisis,” Quarterly Journal of
Economics, Vol. 124, No. 4, November, pp. 1449-1496.
Park, Kwangwoo, and George Pennacchi, 2009,
“Harming depositors and helping borrowers: The
disparate impact of bank consolidation,” Review of
Financial Studies, Vol. 22, No. 1, January, pp. 1-40.
Rosen, Richard J., 2008, “The role of lenders in the
home price boom,” Federal Reserve Bank of Chicago,
working paper, No. WP-2008-16, November.
Case, Karl E., and Robert J. Shiller, 2003, “Is there
a bubble in the housing market?,” Brookings Papers
on Economic Activity’, Vol. 34, No. 2, pp. 299-362.
__________ , 2007a, “Banking market conditions
and deposit interest rates,” Journal ofBanking and
Finance, Vol. 31, No. 12, December, pp. 3862-3884.
Demyanyk, Yuliya, and Otto Van Hemert, 2009,
“Understanding the subprime mortgage crisis,”
Review ofFinancial Studies, May 4, available
at http://rfs.oxfordjoumals.org/content/early/
2OO9/O5/O4/rfs.hhpO33.
__________ , 2007b, “The role of securitization in
mortgage lending,” Chicago Fed Letter, Federal
Reserve Bank of Chicago, No. 244, November.
Federal Financial Institutions Examination
Council, 2008, A Guide to HMDA Reporting:
Getting It Right!, report, Arlington, VA, June, avail
able at www.ffiec.gov/Hmda/pdf/2008guide.pdf.
Gyourko, Joseph, Christopher Mayer, and Todd
Sinai, 2006, “Superstar cities,” National Bureau of
Economic Research, working paper, No. 12355, July,
available at www.nber.org/papers/wl2355.
Haines Cabray L., and Richard J. Rosen, 2007,
“Bubble, bubble, toil, and trouble,” Economic
Perspectives, Federal Reserve Bank of Chicago,
Vol. 31, No. 1, First Quarter, pp. 16-35.
Steverman, Ben, and David Bogoslaw, 2008,
“The financial crisis blame game,” BusinessWeek.com,
October 18, available atwww.businessweek.com/
investor/content/oct2008/pi20081017 9503 82.htm.
U.S. Congress, Joint Economic Committee, 2007,
“The subprime lending crisis: The economic impact
on wealth, property values and tax revenues, and how
we got here,” report, Washington, DC, October, avail
able at www.jec.senate.gov/archive/Documents/
Reports/10.25.07OctoberSubprimeReport.pdf.
Wholesale Access Mortgage Research and
Consulting Inc., 2005, Mortgage Brokers 2004,
report, Columbia, MD.
Inside Mortgage Finance Publications, 2008,
The 2008 Mortgage Market Statistical Annual,
2 vols., Bethesda, MD.
Federal Reserve Bank of Chicago
21
Monitoring financial stability: A financial conditions
index approach
Scott Brave and R. Andrew Butters
Introduction and summary
One of the key observations to come out of the recent
crisis is that financial innovation has made it difficult
to capture broad financial conditions in a small number of
variables covering just a few traditional financial markets.
The network of financial firms outside the traditional
commercial banking system—that is, the so-called shad
ow banking system—was at the forefront of many of
the major events of the crisis, as were newer financial
markets for derivatives and asset-backed securities.
In the wake of the crisis, policymakers, regulators,
financial market participants, and researchers have all
affirmed the importance of the interconnections between
traditional and newly developed financial markets, as
well as their linkages to the nonfinancial sectors of the
economy. The Dodd-Frank Wall Street Reform and
Consumer Protection Act of 2010 sets forth a financial
stability mandate built on this widespread affirmation.
Monitoring financial stability, thus, now explicitly
requires an understanding of both how traditional and
evolving financial markets relate to each other and
how they relate to economic conditions. Indexes of
financial conditions are an attempt to quantify these
relationships. Here, we describe two new indexes that
expand on the work of Illing and Liu (2006), Nelson
and Perli (2007), Hakkio and Keeton (2009), and
Hatzius et al. (2010).
In what follows, we first describe our method
of index construction. The novel contribution of our
method is that it takes into account both the cross
correlations of a large number of financial variables
and the historical evolution of the index to derive a
set of weights for each element of the index. We also
develop an alternative index that separates the influ
ence of economic conditions from financial conditions.
We then highlight the contribution of different sectors
of the financial system to our indexes, as well as the
systemically important indicators among them.
22
Next, we show that the indexes of financial con
ditions we produce are useful tools in gauging finan
cial stability. Major events in U.S. financial history
are well captured by the history of our indexes, as is
the interdependence of financial and economic condi
tions. To further demonstrate the latter, we establish
that it is possible to use our indexes to improve upon
forecasts of measures of economic activity over short
and medium forecast horizons.
Measuring financial conditions
Indexes of financial conditions are typically con
structed as weighted averages of a number of indicators
of the financial system’s health. Commonly, a statisti
cal method called principal component analysis, or
PCA, is used to estimate the weight given each indi
cator (see box 1 for details). The benefit of PCA is
its ability to determine the individual importance of
a large number of indicators so that the weight each
receives is consistent with its historical importance
to fluctuations in the broader financial system.
Indexes of this sort have the advantage of captur
ing the interconnectedness of financial markets—a de
sirable feature allowing for an interpretation of the
systemic importance of each indicator. The more
correlated an indicator is with its peers, the higher the
weight it receives. This allows for the possibility that
a small deterioration in a heavily weighted indicator
may mean more for financial stability than a large
deterioration in an indicator of little weight.
Scott Brave is a senior business economist and R. Andrew
Butters is a former associate economist in the Economic
Research Department at the Federal Reserve Bank ofChicago.
The authors thank Hesna Genay, Spencer Krane, Alejandro
Justiniano, Gadi Barlevy, Jeff Campbell, Douglas Evanoff
and Lisa Barrowfor their helpful comments.
1Q/2011, Economic Perspectives
BOX 1
What is principal component analysis?
Here, we explain the mathematics behind PC A.1
In our explanation, x denotes the 1 x N element row
vector of data at time I. The first step is to form the
stacked matrix of data vectors X , where each column
of this vector contains T observations of a financial
indicator normalized to have a mean of zero and a
standard deviation of one. The eigenvector-eigenvalue
decomposition of the variance-covariance matrix
XTXT then produces a set of weights referenced by
the A x 1 vector W corresponding to the eigenvector
associated with the largest eigenvalue of this matrix.2
These weights are used to construct a weighted sum
of the x at each point in time such that the resulting
index is given by / = A' W.
In a general setting, variation in the frequency or
availability of data makes PCA infeasible. To circum
vent this issue, many indexes restrict the set of financial
indicators and the time period examined at the cost of
losing coverage of more recently developed financial
markets and longer historical comparisons. Alternatively,
Stock and Watson (2002) show how this issue can be
addressed by an iterative estimation strategy that relies
on the incomplete data methods of the expectationmaximization (EM) algorithm of Watson and Engle
(1983). As the number of indicators becomes large,
this strategy produces an index estimate with the same
desirable statistical properties as PCA.
The EM algorithm uses the information from the
complete, or “balanced,” panel of indicators to make
the best possible prediction of the incomplete, or
“unbalanced,” panel of indicators. Stock and Watson’s
(2002) EM algorithm begins with estimation by PCA
The PCA method also has its limitations, however.
For instance, often the choice of which financial indi
cators to include is restricted by the frequency of data
availability, as well as the length of time for which data
are available. Work by Stock and Watson (2002) and
others have shown how to relax some of these constraints,
and we pursue this direction further so as to construct
a richer and longer time series for our indexes.
Our goal is to be able to construct high-frequency
indexes with broad coverage of measures of risk, liquid
ity, and leverage. By risk, we mean both the premium
placed on risky assets embedded in their returns and
the volatility of asset prices. In terms of liquidity, our
measures capture the willingness to both borrow and
lend at prevailing prices. Measures of leverage, in
turn, provide a reference point for financial debt rela
tive to equity.
Federal Reserve Bank of Chicago
on a balanced subset of the data to obtain an initial
estimate of the index. Data for each of the financial
indicators are then regressed on this estimate of the
index, and the results of each regression are used to
predict missing data. The index is then reestimated
by PCA on both the actual and predicted data. This
process continues until the difference in the sum of
the squared prediction errors between iterations
reaches a desired level of convergence.
Stock and Watson’s (2002) EM algorithm is,
however, a purely static estimation method and does
not incorporate information along the time dimension
into the construction of the index. In addition, it, too,
is restricted by the need for an initial balanced panel
of the highest-frequency indicators, given its reliance
on PCA. Because most high-frequency financial indi
cators are not readily available prior to the mid-1980s,
this constraint is not trivial. We, instead, use this
method as a starting point, but then rely on the alter
native estimation procedure ofDoz, Giannone, and
Reichlin (2006). Their method allows us to also in
corporate information along the time dimension into
our index, and is a form of what is referred to as
dynamic factor analysis.
’For further details on PCA, see Theil (1971), pp. 46 48.
Underlying the normalization of the data is the concept of
“stationarity,” or the notion that the mean and variance of
each indicator do not vary over time. For this to be true, some
indicators must first be altered with a stationarity-inducing
transformation prior to estimation. The stationarity-inducing
transformations we used can be found in table A1 in the
appendix.
To allow for historical comparisons and financial
innovation, our method must also be able to incorporate
time series of varying lengths and different frequencies.
To do so, we apply the methods ofDoz, Giannone, and
Reichlin (2006) and Amoba, Diebold, and Scotti (2009)
(see box 2 for details). This framework allows us to
make use of weekly, monthly, and quarterly financial
indicators with histories that potentially begin and
end at different times.
To briefly describe our method, we add a second
dimension to the averaging process—namely, the timeseries dimension of the index. At each point in time,
all of the available indicators are used to construct the
index, ignoring those that are unavailable. The histor
ical time-series dynamics of the index are then used
to smooth its history; and when these indicators again
23
BOX 2
Estimating our financial conditions indexes
Our FCI is constructed in a similar fashion to many
coincident indicator models where the variation in
a panel of time series is governed linearly by an un
known common source and an idiosyncratic error
term. The static measurement equation these models
all have in common is of the following form:
where F represents a 1 x T latent factor capturing a
time-varying common source of variation in the N>'T
matrix of standardized and stationary financial indica
tors and T represents its N x 1 loadings onto this
factor. A defining characteristic of X for our FCI is
its large size in both the cross section (N) and time
domain (7).
Adding dynamics of some finite order to the
latent factor moves the model into the large approxi
mate dynamic factor framework of Doz, Giannone,
and Reichlin (2006). The state-space representation
of this model is given by:
At = TFt + e,
t’
where T are factor loadings estimated off the cross
section of financial indicators and A is the transition
matrix describing the evolution of the latent factor
become available, the history is updated to reflect the
information gained.
Using this method, we construct our weekly finan
cial conditions index (FCI) that takes into account both
the cross-correlations of the indicators and the historical
evolution of the index itself in determining the appro
priate weights. The latter serves to smooth changes to
the index over time, leaving behind more persistent con
tributions from the indicators. This feature is desirable,
particularly in real time, because it avoids putting too
much emphasis on potentially temporary factors influ
encing financial conditions.
Following Hatzius et al. (2010), we also consider
adjusting the indicators for current and past economic
activity and inflation prior to construction of the index.
Our “adjusted” FCI, described in box 2, is motivated
by the observation that financial and economic condi
tions are highly correlated. Removing the variation
explained by the latter addresses potential asymmetries
in the response of one to the other. For instance, a
24
over time. The latent factor’s dynamics,/?, as ex
pressed in the transition matrix A are assumed to be
of finite order: p = 15 weeks. Fifteen lags correspond
roughly with one quarter’s worth of data.
With the model in state-space form and initial
estimates of the system matrices, the EM algorithm
outlined by Shumway and Staffer (1982) can be used
to estimate the latent factor F. At each iteration of
the algorithm, one pass of the data through the Kalman
filter and smoother is made, followed by reestima
tion of the system matrices by linear regression.1 The
log-likelihood function that results is nondecreasing,
and convergence is governed by its stability.
We use the PCA-based EM algorithm of Stock
and Watson (2002) to provide consistent initial estimates
£ £
of r and
and we use linear regression on the
N
subsequent estimate of Fto obtain consistent initial
ft is worth emphasizing,
T
however, that these initializations are more restrictive
than necessary and serve in this framework only to
considerably reduce the required number of iterations
of the EM algorithm. For instance, PCA normalizes
estimates of A and
the factor loadings to satisfy
r'r = I and assumes
that —C1 = a2/. The large approximate dynamic factor
deterioration in financial conditions when economic
growth is high and inflation low may have different
effects on the real economy than a deterioration in
financial conditions when economic growth is low
and inflation high.
Our adjusted FCI is, thus, likely relevant for iso
lating the source of the shock to financial conditions.1
That said, our FCI is a broader metric of financial sta
bility because it captures the interaction of financial
conditions and economic conditions. Combined, the
two indexes could serve as usefiil policy tools by pro
viding a sense of how tight or loose financial markets
are operating relative to historical norms.
Figure 1 plots our FCI and adjusted FCI. Interpreting
the level of both requires a reference to some historical
norm. The norm considered in figure 1 is the sample
mean of each index, which provides a sense of the aver
age state of financial conditions, or its long-term his
torical trend. In this sense, a zero value for our FCI in
figure 1 corresponds with a financial system operating
1Q/2011, Economic Perspectives
BOX
2 (CONTINUED)
Estimating our financial conditions indexes
model framework relaxes this assumption, instead using
vv
the normalization that
= I and accommodating
cross-sectional heteroskedasticity, that is,
Because of the varying frequencies of observa
tion of the data in our FCI, we must also make two
extensions to the EM algorithm prior to estimation.
The first involves setting up the Kalman filter to
deal with missing values as discussed by Durbin and
Koopman (2001). The second modification involves
including additional state variables that evolve deter
ministically to adjust for the temporal aggregation
issues caused by the varying frequencies of data
observation. Here, we follow Aruoba, Diebold, and
Scotti (2009) in their application of Harvey (1989)
to data of varying frequencies of observation to
augment the transition dynamics of the state-space
model accordingly.
Our adjusted FCI requires pretreatment of the data
before application of the routine we just described. Each
of the 100 financial variables is first regressed on current
and lagged values of a measure of the business cycle—
that is, the three-month moving average of the Chicago
Fed National Activity Index (CFNAI-MA3)—and infla
tion—that is, three-month total inflation as measured
by the Personal Consumption Expenditures (PCE)
at the historical average levels of risk, liquidity, and
leverage. For our adjusted FCI, a zero value indicates
a financial system operating at the historical average
levels of risk, liquidity, and leverage consistent with
economic conditions.
In general, risk measures receive positive weights
in each index, whereas liquidity and leverage measures
tend to have negative weights. This pattern of increasing
risk premiums and declining liquidity and leverage is
consistent with tightening financial conditions, and pro
vides us a basis for interpreting both indexes: Positive
index values indicate tighter conditions than on average,
and negative index values indicate looser conditions
than on average.
In addition, it is common for financial conditions
indexes to be expressed relative to their sample standard
deviations. We follow this approach to establish a scale
for our FCI and adjusted FCI in figure 1. Measured in
this way, an index value of 1.0 is associated with finan
cial conditions that are tighter than on average by one
Federal Reserve Bank of Chicago
Price Index—with the number of current and lagged
values in each regression chosen for each variable using
the Bayesian Information Criterion. The independent
variables of these regressions were transformed so as
to match the frequency of observation of the dependent
variable. For weekly variables, we assumed only lagged
values enter the regression and that these values are
constant over the weeks of the month because of the
monthly frequency of observation for the CFNAI-MA3
and total PCE inflation. The standardized residuals
from these regressions are then used to construct our
adjusted FCI.
Our 100 financial indicators consist of 47 weekly,
29 monthly, and 24 quarterly variables. The longest
time series extends back to 1971, while the shortest
begins in 2008. We estimate the EM algorithm on the
unbalanced panel from the first week of 1971 through
2010. However, we only consider the estimates from
the first week of 1973 onward. At this point, over
25 percent of the financial indicators we examine
have complete time series. Because of the number
of high-frequency indicators we examine, it is not
until 1987 that 50 percent have complete time series.
’In addition, a small alteration in the least squares step is re
quired to account for the fact that the unobserved components
of the model must first be estimated. See Brave and Butters
(2010a) for more information on the construction of the index.
standard deviation. Similarly, an index value of-1.0
indicates that financial conditions are looser than on
average by one standard deviation.
It is important to note, however, that given the
transformations described previously, direct compari
sons across the two indexes are not valid. Instead,
comparisons must be made with respect to how each
captures financial conditions over time. For instance,
our adjusted FCI is much less persistent, moving above
and below its average value more frequently than our
FCI. It is also the case that our adjusted FCI gives
more emphasis to recent financial market disruptions,
often putting them on par with the more volatile
1970s and 1980s.
Instances can occur where adjusting for economic
conditions produces a different interpretation of finan
cial conditions than our FCI. Periods of high economic
growth, such as the mid-1980s and late 1990s, often
lead to an above-average adjusted FCI when our
FCI is below average. Conversely, periods of high
25
FIGURE 1
Financial conditions indexes (FCI and adjusted FCI), 1973-2010
1Q /2 011, Econom ic Per spe ctiv es
inflation, such as the 1970s and early 1980s, often
lead to a below-average adjusted FCI when our FCI
is above average.
Systemically important indicators
There are two ways to view the systemic relation
ship expressed in each indicator’s weight: by its sign
and by its magnitude. Risk measures with their gener
ally positive weights and liquidity and leverage mea
sures with their generally negative weights imply that
increasingly positive values of the index capture periods
of above-average risk and below-average liquidity and
leverage. Conversely, increasingly negative values of
the index capture periods where risk premiums are below
average and liquidity and leverage are above average.
The way in which leverage enters our indexes is
in line with Adrian and Shin (2010), who find leverage
is often procyclical (that is, it is positively correlated
with the overall state of the economy). In this way, the
process of deleveraging appears in the indexes as an
indicator of deteriorating financial conditions. Unlike
other methods, however, our estimation framework
can potentially take into account that a buildup of
leverage generates a tendency to reverse itself that
depends on the degree of mean reversion that our
FCI and adjusted FCI have shown over time.
Taking into account the financial markets represent
ed, we have segmented the financial indicators in our
FCI and adjusted FCI into three categories: money
markets (28 indicators), debt and equity markets (27),
and the banking system (45). Table At in the appendix
summarizes all 100 financial indicators in the form they
enter both indexes; the indicators are listed in this
order—from those with the largest positive weights
to those with the largest negative weights within each
category for our FCI. Because in our estimation method
the weights are only identified up to scale, we have
scaled them to have a unit variance in the table for
ease of comparison.
The money markets category is made up mostly
of interest rate spreads that form the basis of most
other financial conditions indexes.2 However, unlike for
many of these indexes, we also include in this category
measures of implied volatility and trading volumes
of several money market financial products. Interest
rate spreads and measures of implied volatility tend
to receive positive weights, whereas trading volumes
tend to receive negative weights. The implication of
this pattern is that widening spreads, increasing vola
tility, and declining volumes all constitute a tightening
in money market conditions.
Some of the interest rate spreads given the great
est positive weights in our FCI include the one-month
Federal Reserve Bank of Chicago
nonfinancial A2P2/AA commercial paper credit spread,
as well as the two-year interest rate swap and the threemonth Libor spreads relative to Treasuries. The first
captures the risk premium for issuing short-term com
mercial paper to less creditworthy borrowers. The re
maining two indicators capture elements of liquidity
and credit risk in the interest rate derivative and inter
bank lending markets, respectively. The Merrill Lynch
implied volatility measures for options and swaptions
(MOVE and SMOVE) also receive large positive
weights, whereas open interest in money market
derivatives and repo market volume receive sizable
negative weights. The former two indicators are, in
a sense, measures of risk, while the latter two can be
viewed as measures of liquidity and leverage.
The debt and equity markets category comprises
mostly equity and bond price measures capturing vol
atility and risk premiums in their various forms. In
addition to stock and bond market prices, we include
in this category residential and commercial real estate
prices, as well as municipal and corporate bond, stock,
asset-backed security, and credit derivative market
volumes. The latter measures capture elements of both
market liquidity and leverage. In general, the indicators
in this category follow the same pattern as the money
market category, so that widening credit spreads, in
creasing volatility, and declining volumes denote
tightening debt and equity market conditions.
In terms of equities, the largest positive weight
in our FCI is given to the Chicago Board Options
Exchange (CBOE) Market Volatility Index, commonly
referred to as the VIX, which measures the implied
volatility of the Standard & Poor’s (S&P) 500; the
largest negative weight is given to the relative valuation
of financial stocks in the S&P 500 (S&P Financials/
S&P 500). In terms of bonds, credit spreads such as
the high yield/Baa corporate and financial/corporate
enter strongly here with large positive weights; so do
spreads relative to Treasuries or swaps for nonmortgage
asset-backed securities (ABS), mortgage-backed
securities (MBS), and commercial-mortgage-backed
securities (CMBS). Swap spreads on credit derivatives
for investment grade and high-yield corporate bonds—
or credit default swaps (CDS), a measure of insurance
protection tied to default—are also given sizable
positive weights.
The banking system category contains mainly
survey-based measures of credit availability as well
as accounting-based measures for commercial banks
and so-called shadow banks, but a few interest rate
spreads also appear in this category. The former indi
cators are primarily measures of liquidity and leverage,
but they also capture the risk tied to deteriorations in
27
FIGURE 2
Decomposition of variance explained by financial conditions indexes
(FCI and adjusted FCI)
^B Money markets
^B Debt and equity markets
Banking system
Note: All values are in percent.
credit quality. Of the interest rate spreads, the difference
between the 30-year jumbo and conforming fixed-rate
mortgages receives the largest positive weight, followed
by the 30-year conforming mortgage/10-year Treasury
yield spread.
The Federal Reserve Board’s Senior Loan Officer
Opinion Survey questions on loan spreads and lending
standards all enter strongly into our FCI (mostly with
large positive weights so that widening spreads and
tighter standards reflect tighter conditions in the banking
system), as do several other survey measures of busi
ness and consumer credit availability. Depending on
how these survey measures are expressed, some receive
large negative weights; but in each case, declining avail
ability coincides with fighter banking system conditions.
The Credit Derivatives Research Counterparty
Risk Index, measured as the average of the CDS spreads
of the largest 14 issuers of CDS contracts, also receives
a large positive weight, with the remaining weight
split roughly evenly between measures of credit quality
and commercial and shadow bank lending and leverage.
All of these measures capture the inherent risks to the
stability of the financial system posed by the potential
collapse of commercial and shadow bank entities.
28
Differences arise in the relative systemic impor
tance of several indicators when considering the impact
of economic conditions in the estimation of the indicator
weights. Figure 2 helps to explain these differences.
Measures of the health of the banking system capture
41 percent of the variation explained by our FCI, fol
lowed by money market measures at 30 percent and
debt and equity market measures at 29 percent. After
performing the same calculation on our adjusted FCI,
we note that money market measures now dominate
at 54 percent, with debt and equity market measures
accounting for 26 percent and the banking system
measures accounting for 20 percent.
Thus, the primary effect of adjusting for economic
conditions appears to be the reduction in importance of
banking system measures. The survey-based indicators
within the banking system category, in particular, show
the largest declines in weight. A lower weight in this
case indicates that much of the variation in these indi
cators can be explained by changes in either economic
activity or inflation over time. A secondary effect, visible
in table Al in the appendix, is the addition of weight
to certain measures of liquidity and leverage—that is,
corporate bond and asset-backed security issuance, the
net notional value of credit derivatives, and several
commercial and shadow bank leverage measures.
1Q/2011, Economic Perspectives
It is likely that some of this result, shown in figure 2,
stems from the fact that most of the previously men
tioned measures are available at a weekly frequency
Our adjustment for economic conditions is more like
ly to account for medium-frequency rather than highfrequency variation. However, an examination of the
weights in table Al suggests that this cannot be the
sole explanation. Several weekly money market mea
sures receive greater weight—for example, the threemonth London interbank bid (Eurodollar) and offered
(TED) rate spreads; but there are also a number of
weekly debt and equity market measures that receive
less—for example, the high yield/Baa corporate bond,
CMBS, and various credit derivative swap spreads, as
well as the VIX.
Gauging financial stability
One way to judge the validity of our indexes as
measures of financial stability is to follow the narra
tive approach and link their values to significant events
in U.S. financial history. To illustrate this point, we
plot our FCI and adjusted FCI in figure 3, highlight
ing prominent historical events.3 Each panel of figure 3
depicts a decade of the index. Events are labeled with
text boxes and arrows directed toward a specific week
of both indexes denoted by a circle marker.
Overall, significant periods of crisis in financial
history are well captured by both indexes, as are periods
of relative calm. There are subtle differences, however,
between the indexes around the time of several of the
major events marked in figure 3. The first is clearly
seen in panel A of figure 3 during the 1973-75 period
that saw disruptions in equity markets and the failures
of several large banks. In general, our adjusted FCI is
quicker than our FCI to note both the onset and end
of pressures—as financial conditions began to deteri
orate prior to the 1973-75 recession and as they be
gan to recover sooner than the real economy.
For most of the rest of the 1970s, both indexes
indicate very similar financial conditions. However,
by the end of the decade and into the early 1980s, as
shown in panels A and B of figure 3, differences again
emerge. The large swings in economic activity and
inflation during these periods lead the adjusted FCI to
be much more volatile, often swinging from well be
low zero to well above it very quickly. At their peak
levels, both indexes are still very similar, capturing
very well the major events of this period.
From the mid-1980s through the end of the decade,
differences between the two indexes are much smaller
(panel B of figure 3). Two events, however, stand out
during this period of strong growth and disinflation:
the resolution of Continental Illinois National Bank
Federal Reserve Bank of Chicago
and Trust Company and the “Black Monday” stock
market crash of 1987; the adjusted FCI puts more
weight relative to earlier events on each compared
with the FCI. The adjusted FCI is also quicker to note
above-average tightness in response to the U.S. savings
and loan crisis and quicker to recover from the crisis
after accounting for the 1990-91 recession (see panels B
and C of figure 3).
From the mid-1990s through the end of the decade
(panel C of figure 3), the adjusted FCI consistently
indicates financial conditions relative to economic con
ditions either about average or tighter than on average.
In contrast, only after the Russian debt default, the sub
sequent collapse of Long-Term Capital Management,
and the run-up to Y2K (the year 2000 software problem)
does the FCI indicate financial conditions that are tighter
than on average. During this period, the adjusted FCI
additionally picks up the relative tightening in financial
markets surrounding the Mexican peso devaluation and
Asian financial crisis (around the time of the devalua
tion of the Thai baht).
Despite small differences surrounding the crash
of the NASDAQ Stock Market and the corporate
accounting scandals of the early 2000s (panel D of
figure 3), both indexes generally indicated conditions
looser than on average through the early part of the
previous decade. Beginning in late 2005, the adjusted
FCI moved closer to its average, while the FCI remained
well below its average. The recent financial crisis
appears at about the same time in both indexes, from
mid-2007 through mid-2009, while the recovery reg
isters a little later in the adjusted FCI.
More recently, as seen in figure 1 (p. 26), both
indexes demonstrate that the financial system has
healed significantly. Financial conditions by either
measure, however, remain tighter than they were be
fore the crisis. They have also been responsive to the
European sovereign debt concerns that began in the
spring of 2010 and the slowdown in economic activity
throughout the summer months of 2010. In fact, our
adjusted FCI breached its average level in the summer
of 2010 before easing again during the rest of 2010.
Our historical analysis shows that persistent de
viations in the interpretation of our two indexes con
tain useful information. The adjusted FCI is, in some
sense, a forward-looking indicator of the FCI. When
financial conditions are out of balance with economic
conditions for an extended period, a correction in the
FCI tends to result. Whether or not this result is due
to the influence of the policy actions taken during
these periods or other economic forces is beyond the
scope of the analysis here. However, we refer the
reader to Brave and Butters (2010a) and Brave and
29
FIGURE 3
Financial conditions indexes (FCI and adjusted FCI) and prominent historical events, 1973-2009
A. 1970s
standard deviations from trend
1Q /2 011, Econom ic Per spe ctiv es
1973
Federa l Res erv e Bank of Chicag o
FIGURE 3 (continued)
Financial conditions indexes (FCI and adjusted FCI) and prominent historical events, 1973-2009
B.1980s
standard deviations from trend
FIGURE 3 (continued)
Financial conditions indexes (FCI and adjusted FCI) and prominent historical events, 1973-2009
C. 1990s
standard deviations from trend
6
4
-2 -
1Q /2 011, Econom ic Per spe ctiv es
-4
1990
’91
’92
’93
’95
’94
FCI
’96
Adjusted FCI
’97
’98
’99
Federa l Res erv e Bank of Chicag o
FIGURE 3 (continued)
Financial conditions indexes (FCI and adjusted FCI) and prominent historical events, 1973-2009
D. 2000s
standard deviations from trend
_____
FCI
...
Adjusted FCI
Genay (2011) for more rigorous analyses of the FCI
and adjusted FCI.
Forecasting economic conditions
Another test of our indexes is their ability to pre
dict the impact of changes in financial conditions on
future economic activity. We follow the forecasting
framework of Hatzius et al. (2010); but we refine their
approach in two ways: 1) by concentrating on the por
tion of our FCI that cannot be explained by its historical
dynamics and 2) by including as explanatory variables
high-frequency nonfinancial measures of economic
activity, such as the Chicago Fed National Activity
Index (CFNAI).4
We refer to the portion of our FCI that cannot be
predicted based on its historical dynamics as the FCI
residual. The FCI residual corresponds with the error
term, v(, from the transition equation of our dynamic
factor model (described in detail in box 2), where we
follow the convention described previously for our FCI
and scale it by its sample standard deviation. Because
the FCI captures an element of financial conditions
that also depends on economic conditions, systematic
changes in the FCI over time reflect the historical
response of financial conditions to past changes in
financial and economic conditions. The FCI residual,
therefore, reflects the deviation of financial conditions
from this historical pattern.
It is this aspect of the FCI residual that we find
appealing as an explanatory variable for future economic
activity; in this regard, we prefer the FCI residual over
the adjusted FCI, which captures only the deviation
of financial conditions from economic conditions.
Hatzius et al. (2010) frame the use of their adjusted
index as a method of focusing purely on the impact
of financial shocks on economic activity. We, instead,
use our FCI because it also contains information on
economic shocks. We then control for whether this
information is in addition to that found in high-frequency
nonfinancial measures of economic activity.
To demonstrate the ability of the FCI residual to
predict future economic conditions and for the sake
of comparison with the adjusted FCI, we conducted a
pseudo out-of-sample forecasting exercise. Our mixedfrequency forecasting regressions incorporated lagged
values of quarterly forecast variables taken from the
U.S. Bureau of Economic Analysis’s national income
and product accounts (NIPA), as well as current and
lagged values of the three-month moving average of
the CFNAI alone or in combination with the 13-week
moving average of one of the following sampled at the
end of each month: the FCI residual, adjusted FCI, or
34
adjusted FCI residual (which is the portion of the ad
justed FCI unexplained by its historical dynamics).5
The CFNAI’s three-month moving average serves
as our reference point in evaluating the marginal in
formation content of our measures of financial condi
tions over high-frequency nonfinancial measures of
economic activity. It is a summary measure of 85 in
dicators constructed using PCA on data for production
and income; employment, unemployment, and hours;
personal consumption and housing; and sales, orders,
and inventories.6 The CFNAI has been used in the
past to forecast economic growth and inflation by
Stock and Watson (1999) and Brave and Butters
(2010b), among others.
Our forecasting regression takes the following
form:
Y,+l~ Y,= « + t PM+1_,.+f yyCFW4/,+1_y
i=l
7=1
+Z SkFCIl+i_k +s,+h,
7=1
where T refers to the natural log of a particular NIPA data
series, CFNAI indicates the three-month moving average
of the CFNAI, and FCI is the 13-week moving average
of either the FCI residual, adjusted FCI, or adjusted FCI
residual. The explanatory variables were aligned with the
NIPA data in the last month of each quarter (t) to match
frequencies so that the index / represents a quarter (or three
months) and the indexes j and k both represent months.
To construct forecasts, we began with data from
1973 :Q 1 through 1984:Q4.7 One quarter’s worth of
data was then added on a recursive basis and forecasts
made at a horizon (//) of one, two, four, and six quarter(s)
ahead until the end of our data in 2010:Q2. The advan
tage of this framework is that it mimics the production
of forecasts in real time (minus the impact of data
revisions). In this way, we can account for model
uncertainty. To allow for the further possibility of a
change in lag structure over time, we had each recur
sive regression incorporate the Bayesian Information
Criterion lag selection method.8
For an evaluation criterion, we used the meansquared forecast error (MSFE) statistic computed from
our sample of forecasts from 1985;Q1 through 2010:Q2
expressed relative to the similar statistic based on fore
casts computed using only lagged quarterly growth rates
of the NIPA variables. This ratio provides a test of model
fit, so that a value less than 1 indicates an improvement
in forecast accuracy relative to an autoregressive base
line for each NIPA variable. The MSFE statistic summa
rizes two elements in our pseudo out-of-sample context:
the improvement in fit from incorporating the CFNAI
1Q/2011, Economic Perspectives
TABLE 1
Pseudo out-of-sample relative MSFE ratios
h
CFNAI
FCI
residual
Adjusted
FCI
Adjusted
FCI residual
0.81
0.82
0.90
0.88
0.88
1.06
1.07
1.07
0.85
0.96
1.00
1.01
0.91
0.88
0.94
1.02
1.03
1.06
1.17
1.20
0.96
0.97
1.10
1.11
0.76
0.67
0.75
0.79
0.79
0.81
0.90
0.89
0.78
0.73
0.85
0.84
0.92
0.99
1.23
1.30
1.11
1.19
1.29
1.37
1.13
1.18
1.33
1.35
1.03
0.97
0.94
0.97
1.10
1.01
0.98
1.03
1.07
1.01
0.99
1.02
A. Gross domestic product
1
2
4
6
0.88
0.98
1.05
1.06
1.06
1.07
1.16
1.18
0.78
0.76
0.86
0.91
1.13
1.18
1.32
1.33
Adjusted
FCI
Adjusted
FCI residual
1
2
4
6
1.06
1.14
1.14
1.17
0.98
0.90
0.98
1.05
1.01
1.14
1.15
1.19
1.00
1.06
1.08
1.11
1
2
4
6
0.59
0.37
0.47
0.64
0.58
0.37
0.40
0.56
0.58
0.37
0.46
0.63
0.60
0.37
0.44
0.61
0.93
1.19
1.11
1.03
0.96
1.00
1.07
1.01
1.00
1.15
1.05
1.07
0.91
0.98
0.98
1.00
F. Residential investment
G. PCE: Durables
1
2
4
6
FCI
residual
D. Nonfarm private inventories
E. Nonresidential investment
1
2
4
6
CFNAI
B. Gross domestic purchases
C. Final sales
1
2
4
6
h
1
2
4
6
1.13
1.17
1.06
1.01
0.92
0.91
0.97
0.95
H. PCE: Nondurables
1
2
4
6
0.95
1.02
1.00
1.03
0.87
0.87
0.89
0.94
I. PCE: Services
1
2
4
6
1.12
1.01
1.01
1.00
Notes: The table displays mean-squared forecast error (MSFE) ratios expressed relative to an autoregressive baseline model. The forecasted
variable is listed at the top of each panel. Column headings for each panel denote the additional variable added to the baseline model: The
CFNAI is the three-month moving average of the Chicago Fed National Activity Index and is included in all four specifications. The FCI residual is
the 13-week moving average of the portion of the financial conditions index unexplained by its historical dynamics, the adjusted FCI is the 13-week
moving average of the financial conditions index adjusted for economic conditions, and the adjusted FCI residual is the 13-week moving average
of the portion of the adjusted financial conditions index unexplained by its historical dynamics—these three individually serve to augment the model
including the CFNAI. The rows in each panel denote the forecast horizon (/?) measured in quarters beyond the end of the sample period. The sample
period is recursive beginning in 1973:Q1 and rolling forward from 1985:Q1 through 2010:Q2. PCE denotes personal consumption expenditures.
Source: Authors’ calculations based on data from the U.S. Bureau of Economic Analysis, National Income and Product Accounts of the United States,
from Haver Analytics.
alone or from incorporating the CFNAI along with
the FCI residual, adjusted FCI, or adjusted FCI residual
to the forecasting regression, balanced against the added
parameter uncertainty from estimating additional re
gression coefficients.
Table 1 summarizes the results for nine NIPA
variables all expressed in real, or constant price, terms.
Gross domestic product (GDP) in panel A is the broad
est measure we consider, but we also examine several
of its components. Gross domestic purchases (panel B)
exclude exports, and thus solely capture domestic de
mand. Final sales (panel C) remove the influence of
changes in inventories. Nonfarm private inventories,
nonresidential investment, and residential investment
(panels D, E, and F) form the basis of the investment
Federal Reserve Bank of Chicago
component of GDP we consider, and personal expendi
tures on durables, nondurables, and services (panels G,
H, and I) account for consumption. We do not directly
consider government spending or exports.
A few observations are readily apparent from this
table. First, including the CFNAI in our forecasting
regressions on NIPA data results in a substantial im
provement in forecast accuracy (MSFE ratios less than
1) for GDP and measures of business investment, par
ticularly at shorter horizons. Adding the FCI residual
improves upon these initial forecasts at every horizon
and for every variable, with the magnitude of improve
ment ranging from just less than 1 percent to 22 percent.9
In contrast, adding the adjusted FCI rarely improves
on the forecasts based on the CFNAI alone; and the
35
FIGURE 4
Two-quarter-ahead forecasts of real gross domestic product growth
1Q /2 011, Econom ic Per spe ctiv es
——— Chicago Fed National Activity Index’s
three-month moving average (CFNAI)
• ••••• CFNAI and financial conditions index (FCI)
residual’s 13-week moving average
---------CFNAI and adjusted FCI residual’s
13-week moving average
Actual
Source: Authors’ calculations based on data from the U.S. Bureau of Economic Analysis, National Income and Product Accounts of the United States, from Haver Analytics.
forecasts augmented with the adjusted FCI are less
accurate than the forecasts augmented with the FCI
residual in nearly every case.
The FCI-residual forecasts are also superior
when compared with the adjusted-FCI-residual fore
casts in nearly every case. However, the adjusted-FCIresidual forecasts are often superior to the forecasts
based on the CFNAI alone and those augmented with
the adjusted FCI. In this respect, our results suggest
how to improve the ability of the adjusted FCI to fore
cast future economic activity—the key is to focus on
the portion of the adjusted FCI that is not explained
by its historical dynamics. This potential improve
ment is made by our extension of the index con
struction methodology of Hatzius et al. (2010) to
a dynamic framework.
The results in table 1 also suggest that the FCI
residual contains information on future economic
activity in addition to that found in high-frequency
nonfinancial measures of economic activity. There
is, however, considerable variation in the forecasting
performance of the FCI residual over time not shown
in table 1. Much of the gains in forecast accuracy are
concentrated in the recent period. Despite this fact,
the inclusion of the FCI residual in our forecasting
regressions rarely significantly worsens the forecast
based on the CFNAI alone, so that it comes with little
cost but potentially large benefits.
Figure 4 captures an instance of the small cost,
large reward nature of including the FCI residual in
our forecasting regression. It depicts actual real GDP
growth at a two-quarter horizon and the forecasts for
this measure based on the CFNAI’s three-month moving
average, as well as these forecasts including the 13-week
moving average of the FCI residual or adjusted FCI
residual. Differences prior to the recent crisis tend to
be small. During these periods, sometimes the forecast
including the FCI residual is marginally superior and
sometimes it is not.
Federal Reserve Bank of Chicago
The forecast series begin to consistently deviate
from one another in the second half of 2007, when
the crisis started to unfold. Throughout the recent
recession and recovery, the forecast including the FCI
residual has more consistently tracked actual real GDP
growth than any of the other forecasts we consider. At
times during this period, however, the adjusted-FCIresidual forecast has been superior. The FCI-residual
forecast’s dominance over the adjusted-FCI-residual
forecast over the entire period is due in large part to it
more quickly picking up the beginning of the recent re
cession and the magnitude of the subsequent recovery.
Conclusion
Our newly constructed financial conditions in
dexes can serve as tools for both policymakers and
financial market participants in gauging the current
state of financial markets. Computed over a long time
horizon and from a large sample of financial indicators
of different frequencies, these indexes provide a time
ly assessment of how tightly or loosely financial mar
kets are operating relative to historical financial and
economic conditions.
As a measure of financial stability, our indexes
exhibit several essential characteristics. Known periods
of financial crisis correspond closely with peak periods
of tightness in each index, and the turning points of
each index coincide with well-known events in U.S.
financial history. Furthermore, our indexes contain
information on future economic activity beyond that
found in nonfinancial measures of economic activity.
Our indexes are also unique in that they derive
from an estimation method that captures both the
systemic importance of traditional and new financial
markets and the dynamic evolution of overall finan
cial conditions. In the future, we plan to develop this
framework further in order to better understand the
channels through which changes in financial condi
tions affect economic activity.
37
NOTES
‘Hatzius et al. (2010) also construct a similar version of their index
of financial conditions and relate it to changes in the federal funds
rate over time. We have found very similar results to theirs; our
adjusted FCI is significantly correlated with measures of monetary
policy, though we have not documented this here. See Brave and
Genay (2011), who relate monetary policy during the recent crisis
to the adjusted FCI, for more information.
2Most of our 100 financial indicators have become standard fare in
the financial press as a result of the recent financial crisis. Rather
than describe each in further detail, we refer interested readers to
the useful summaries found in Nelson and Perli (2007), Hakkio and
Keeton (2009), and Hatzius et al. (2010).
3The literature on financial crises is quite extensive. The following
works are a few of those that were instrumental in constructing our
timeline of events: Federal Deposit Insurance Corporation (1984,
1997), Reinhart and Rogoff (2008), Schreft (1990), Minsky (1986),
Spero (1999), Laeven and Valencia (2008), Carron (1982), and
El-Gamal and Jaffe (2008).
5We use smoothed measures of the explanatory variables when
appropriate to approximate the quarterly frequency of the NIPA
variables being forecasted.
6For more details on the CFNAI, including its 85 indicators, see
www.chicagofed.org/digital_assets/publications/cfnai/background/
cfnai_background.pdf.
7To be technically correct, we varied the endpoint of the initial
sample based on the forecast horizon so that the first forecast
always began at 1985:Q1.
8Maximums of I = 5 quarters and J,K= 6 months were used in
its calculation.
9In the case of nonfarm private inventories, there is one instance
in table 1 where the improvement is not apparent because of the
rounding in this table.
4Hakkio and Keeton (2009) also use the CFNAI to make similar
comparisons.
38
1Q/2011, Economic Perspectives
Financial indicator
Transformation
Frequency
Haver/Bloomberg*/
Call ReportA mnemonic
Start
Category
FCI
Adjusted
FCI
1-month Nonfinancial CP A2P2/AA credit spread
LV
W
FAP1M-FCP1M
1997w2
1
2.255
2.308
2-year Swap/Treasury yield spread
3-monthTED spread (Libor-Treasury)
LV
LV
W
T111W2-R111G2
FLOD3-FTBS3
1987W14
1980W23
1
1
2.229
1.825
2.975
3.606
1-month Merrill Lynch Options Volatility Expectations (MOVE)
3-month Merrill Lynch Swaption Volatility Expectations (SMOVE)
LV
LV
SPMLV1
SPMLSV3
1988W15
1996w49
1
1
1.690
1.678
1.566
0.564
3-month/1-week AA Financial CP spread
1-month Asset-backed/Financial CP credit spread
LV
LV
FFP3M-FFP7D
FAB1M-FFP1M
1997w2
2001 w1
1
1
1.582
1.581
2.037
2.064
3-month Eurodollar spread (LIBID-Treasury)
On-the-run vs. Off-the-run 10-year Treasury liquidity premium
LV
LV
FDB3-FTBS3
FYCEPA-FCM10
1971 w2
1985w1
1
1
1.522
0.974
3.048
0.916
10-year Swap/Treasury yield spread
3-month Financial CP/Treasury bill spread
LV
LV
T111WA-R111GA
FFP3-FTBS3
1987W14
1971W1
1
1
0.845
0.619
1.189
1.741
Fed Funds/Overnight Treasury Repo rate spread
3-month OIS/Treasury yield spread
LV
LV
FFED-RPGT01D*
T111W3M-R111G3M
1991 w21
2003W38
1
1
0.495
0.452
1.084
1.352
FDDM/(FDDM+FDTM)
FLOD1Y-FLOD1
1994W40
1986w2
1
1
0.426
0.368
0.430
0.378
FDDG/(FDDG+FDTG)
FDDS/(FDDS+FDTS)
1994W40
1994w40
1
1
0.307
0.168
0.474
0.045
FFED-RPAG01D*
FDDC/(FDDC+FDTC)
1991 w21
2001 w40
1
1
0.150
0.103
0.592
0.051
FFED-RPMB01D*
FCM10
1991 w21
1971w2
1
1
0.037
-0.050
0.173
-0.208
SPMD
RPGT03M*-RPGT01W*
1971 w5
1991W21
1
1
-0.122
-0.141
-0.203
0.858
FYCEP2-FTBS3
FCPT
1971w1
1995W45
1
1
-0.237
-0.482
0.167
-0.231
FYCEPA-FYCEP2
1971 w34
1
-0.706
-0.979
2002w7
1994w40
1
1
-1.024
-1.331
-0.075
-1.078
Agency MBS Repo Delivery Failures Rate
1-year/1-month Libor spread
DLNQ
LV
Treasury Repo Delivery Failures Rate
Agency Repo Delivery Failures Rate
DLNQ
DLNQ
Fed Funds/Overnight Agency Repo rate spread
Corporate Securities Repo Delivery Failures Rate
LV
DLNQ
Fed Funds/Overnight MBS Repo rate spread
10-year Constant Maturity Treasury yield
LV
DLV
Broker-dealer Debit Balances in Margin Accounts
3-month/1-week Treasury Repo spread
DLN
LV
2-year/3-month Treasury yield spread
Commercial Paper Outstanding
LV
DLN
10-year/2-year Treasury yield spread
3-month Eurodollar, 10-year/3-month swap, 2-year and
10-year Treasury Options and Futures Open Interest
Total Repo Market Volume (Repurchases+Reverse Repurchases)
w
w
w
w
w
w
w
w
w
w
w
w
w
w
w
w
w
w
w
M
LV
w
w
w
w
DLNQ
DLNQ
w
w
COPED3P+COPTN2P+COPT10P+COPIRSP
FDFR+FDFV
Citigroup Global Markets ABS/5-yearTreasury yield spread
LV
M
SYCAAB-FCM5
1989W52
2
2.487
2.865
Bloomberg 5-year AAA CMBS spread to Treasuries
Merrill Lynch High Yield/Moody’s Baa corporate bond yield spread
LV
LV
CMBSAAA5*
FMLHY-FBAA
1996W27
1997w2
2
2
2.234
2.116
1.647
0.659
CBOE S&P 500 Volatility Index (VIX)
Credit Derivatives Research North America Investment Grade Index
LV
LV
1990w1
2006w1
2
2
2.074
1.528
1.815
0.477
Credit Derivatives Research North America High Yield Index
Citigroup Global Markets Financial/Corporate Credit bond spread
LV
LV
w
w
w
w
w
Citigroup Global Markets MBS/10-year Treasury yield spread
Bond Market Association Municipal Swap/20-year Treasury yield spread
20-yearTreasury/State & Local Government 20-year General
Obligation Bond yield spread
SPVIX
S009LIG
M
S009LHY
SYCF-SYCT
2006w1
1979W52
2
2
1.516
1.179
0.495
1.959
LV
LV
M
W
SYMT-FCM10
SBMAS-FCM20
1979W52
1989W27
2
2
0.848
0.818
1.568
1.561
LV
W
FSLB-FCM20
1971W1
2
0.502
-0.189
APPENDIX
Federa l Res erv e Bank of Chicag o
TABLE A1
Financial indicators in the financial conditions indexes (FCI and adjusted FCI)
TABLE A1 (continued)
Financial indicators in the financial conditions indexes (FCI and adjusted FCI)
Financial indicator
Moody’s Baa corporate bond/10-year Treasury yield spread
Total Money Market Mutual Fund Assets/Total Long-term Fund Assets
Transformation
Frequency
Haver/BI oom berg*/
Call ReportA mnemonic
Start
Category
FCI
Adjusted
FCI
LV
LV
W
M
FBAA-FCM10
ICMMA/ICIA
1971w1
1974W52
2
2
0.348
0.231
0.936
0.177
DLN
DLN
Q
Q
XL14TCRE5/GDP
(XL31CRE5+XL21TCR5)/GDP
1971W13
1971W13
2
2
0.025
0.024
0.091
0.010
Total MBS Issuance (Relative to 12-month MA)
S&P 500, NASDAQ, and NYSE Market Capitalization/GDP
LVMA
DLN
M
Q
N/A
(SPSP5CAP+SPNYCAPH+SPNACAP)/GDP
2000W52
1971W13
2
2
-0.022
-0.041
-0.106
-0.079
New US Corporate Equity Issuance (Relative to 12-month MA)
Wilshire 5000 Stock Price Index
LVMA
DLN
M
M
FNSIPS
SPWIE
1987W52
1971w5
2
2
-0.047
-0.052
0.027
-0.108
DLN
LV
M
M
USLPHPIS
FNSIS
1976w9
2004w9
2
2
-0.066
-0.108
-0.146
-0.185
Nonfinancial business debt outstanding/GDP
Federal, state, and local debt outstanding/GDP
Loan Performance Home Price Index
New State & Local Government Debt Issues (Relative to 12-month MA)
MIT Center for Real Estate Transactions-Based Commercial Property
Price Index
1Q /2 011, Econom ic Per spe ctiv es
DLN
Q
MTBIP
1984W26
2
-0.111
-0.128
Nonmortgage ABS Issuance (Relative to 12-month MA)
S&P 500, S&P 500 mini, NASDAQ 100, NASDAQ mini Options and Futures
Open Interest
LVMA
M
N/A
2000w52
2
-0.130
-0.184
DLNQ
W
COPSPMP+COPSP5P+COPNAMP+COPNASP
2000W12
2
-0.134
-0.250
CMBS Issuance (Relative to 12-month MA)
New US Corporate Debt Issuance (Relative to 12-month MA)
LVMA
LVMA
M
M
N/A
FNSIPB
1990W52
1987W52
2
2
-0.157
-0.179
-0.184
-0.279
Net Notional Value of Credit Derivatives
S&P 500 Financials/S&P 500 Price Index (Relative to 2-year MA)
DLN
LVMA
W
W
D001TOTH
S5N40I/SPN5COM
2008W45
1989W37
2
2
-0.256
-1.860
-0.522
-2.007
Sr Loan Officer Opinion Survey: Tightening standards on Small C&l Loans
Sr Loan Officer Opinion Survey: Increasing spreads on Small C&l Loans
LV
LV
Q
Q
FTCIS
FSCIS
1990W13
1990W13
3
3
2.501
2.467
1.366
1.312
Sr Loan Officer Opinion Survey: Tightening standards on CRE Loans
Sr Loan Officer Opinion Survey: Tightening standards on Large C&l Loans
LV
LV
Q
Q
FTCRE
FTCIL
1990W26
1990W13
3
3
2.418
2.416
1.442
1.274
Sr Loan Officer Opinion Survey: Increasing spreads on Large C&l Loans
30-year Jumbo/Conforming fixed-rate mortgage spread
LV
LV
Q
W
FSCIL
ILMJNAVG*-ILM3NAVG*
1990W13
1998W23
3
3
2.364
2.220
1.060
2.078
Credit Derivatives Research Counterparty Risk Index
National Federation of Independent Business Survey: Credit Harder to Get
LV
LV
W
M
S000CRI
NFIB20
2006w1
1973W44
3
3
1.361
1.228
0.644
0.668
30-year Conforming Mortgage/10-year Treasury yield spread
American Bankers Association Value of Delinquent Home Equity Loans/
Total Loans
LV
W
FRM30F-FCM10
1978W35
3
1.154
1.491
DLV
M
USHWODA
1999w9
3
0.284
0.169
American Bankers Association Value of Delinquent Consumer Loans/
Total Loans
DLV
M
USSUMDA
1999w9
3
0.264
0.106
1999w9
3
0.220
0.090
1992w9
1984W26
3
3
0.157
0.139
0.024
0.146
1999w9
1973w9
3
3
0.139
0.068
0.197
0.191
1972W26
3
0.028
0.078
American Bankers Association Value of Delinquent Credit Card Loans/
Total Loans
DLV
M
USBKCDA
S&P US Credit Card Quality Index 3-month Delinquency Rate
Noncurrent/Total Loans at Commercial Banks
DLV
DLN
M
Q
CCQID3
(RCFD1407A+RCFD1403A)/RCFD2122A
DLV
DLNQ
M
W
USREVDA
FABWCA/FAA
DLV
Q
USL14FA+USL149A
American Bankers Association Value of Delinquent Non-card Revolving
Credit Loans/Total Loans
Commercial Bank C&l Loans/Total Assets
Mortgage Bankers Association Serious Delinquencies
APPENDIX (continued)
o
Financial indicator
Transformation
Frequency
Haver/Bloomberg*/
Call Report* mnemonic
Start
Category
FCI
Adjusted
FCI
Total Assets of Funding Corporations/GDP
Mortgage Bankers Association Mortgage Applications Volume Market Index
DLN
DLN
Q
W
OA50TAO5/GDP
MBAM
1971W13
1990w2
3
3
0.022
0.020
0.022
-0.086
Total Assets of Agency and GSE backed mortgage pools/GDP
Total Assets of ABS issuers/GDP
DLN
DLN
Q
Q
OA41MOR5/GDP
OA67TAO5/GDP
1971W13
1983W39
3
3
0.011
0.005
0.031
0.025
FDIC Volatile Bank Liabilities
DLN
Q
RCON2604A+RCFN2200A+RCFD2800A
+MAX(RCFD2890A,RCFD3190A)+RCFD3548A
1978W26
3
0.000
0.017
DLNQ
W
FBDA/FAA
1973w9
3
0.000
-0.026
Commercial Bank Deposits/Total Assets
Fed funds and Reverse Repurchase Agreements w/ nonbanks and
Interbank Loans/Total Assets
DLNQ
W
(FAIFFA+FABWORA)/FAA
1973w9
3
-0.005
-0.060
Total Assets of Finance Companies/GDP
Total Unused C&l Loan Commitments/Total Assets
DLN
DLN
Q
Q
OA61TAO5/GDP
RCON3423A/RCON2170A
1971W13
1984W26
3
3
-0.009
-0.011
0.012
-0.036
Total REIT Assets/GDP
Total Assets of Broker-dealers/GDP
DLN
DLN
Q
Q
OA64TAO5/GDP
OA66TAO5/GDP
1971W13
1971W13
3
3
-0.012
-0.013
0.071
-0.035
DLNQ
DLN
W
Q
FABWRA/FAA
OA57TAO5/GDP
1973w9
1971W13
3
3
-0.019
-0.023
-0.026
-0.053
MZM Money Supply
Total Assets of Insurance Companies/GDP
DLN
DLN
M
Q
FMZM
(OA51 TAO5+OA54TAO5)/G DP
1974w9
1971W13
3
3
-0.028
-0.029
-0.076
-0.067
Commercial Bank48-month New Car Loan/2-year Treasury yield spread
Consumer Credit Outstanding
LV
DLN
Q
M
FK48NC-FCM2
FOT
1976W26
1971 w5
3
3
-0.033
-0.039
-0.135
0.057
Commercial Bank Securities in Bank Credit/Total Assets
Commercial Bank 24-month Personal Loan/2-year Treasury yield spread
DLNQ
LV
W
Q
FABYA/FAA
FK24P-FCM2
1973w9
1976W26
3
3
-0.052
-0.083
-0.159
-0.172
S&P US Credit Card Quality Index Receivables Outstanding
S&P US Credit Card Quality Index Excess Rate Spread
DLN
LV
M
M
CCQIO
CCQIX
1992w9
1992w5
3
3
-0.095
-0.109
-0.013
-0.387
Finance Company Receivables Outstanding
Finance Company New Car Loan interest rate/2-year Treasury yield spread
DLN
LV
M
M
FROT
FFINC-FCM2
1985W31
1976W26
3
3
-0.149
-0.150
0.041
-1.130
Sr Loan Officer Opinion Survey: Willingness to Lend to Consumers
UM Household Survey: Auto Credit Conditions Good/Bad spread
LV
LV
Q
M
FWILL
N/A
1971W13
1978w5
3
3
-0.538
-1.354
-0.334
-1.321
UM Household Survey: Mortgage Credit Conditions Good/Bad spread
UM Household Survey: Durable Goods Credit Conditions Good/Bad spread
LV
LV
M
M
N/A
N/A
1978w5
1978w5
3
3
-1.487
-1.543
-1.802
-1.668
National Association of Credit Managers Index
LV
M
CMI
2002w9
3
-2.004
-0.130
Commercial Bank Real Estate Loans/Total Assets
Total Assets of Pension Funds/GDP
Transformations
LV: Level
LVMA: Level relative to moving average
DLV: First difference
DLN: Log first difference
DLNQ: 13-week log difference
Categories
1. Money markets
2. Debt and equity markets
3. Banking system
Notes: All of the financial indicators are in basis points or percentages. N/A means not applicable; the relevant series are taken from Inside Mortgage Finance Publications, CRE Finance Council, and University
of Michigan data. For more information on the indicators, please contact the authors.
APPENDIX (continued)
Federa l Res erv e Bank of Chicag o
TABLE Al (continued)
Financial indicators in the financial conditions indexes (FCI and adjusted FCI)
REFERENCES
Adrian, T., and H. S. Shin, 2010, “Liquidity and
leverage,” Journal ofFinancial Intermediation,
Vol. 19, No. 3, July, pp. 418-437.
Aruoba, S. B., F. X. Diebold, and C. Scotti, 2009,
“Real-time measurement of business conditions,”
Journal ofBusiness and Economic Statistics, Vol. 27,
No. 4, pp. 417-427.
Brave, S., and R. A. Butters, 2010a, “Gathering
insights on the forest from the trees: Anew metric for
financial conditions,” Federal Reserve Bank of Chicago,
working paper, No. WP-2010-07, August 24.
__________ , 2010b, “Chicago Fed National Activity
Index turns ten—Analyzing its first decade of perfor
mance,” Chicago Fed Letter, Federal Reserve Bank
of Chicago, No. 273, April.
Brave, S., and H. Genay, 2011, “Federal Reserve
policies and financial market conditions during the
crisis,” Federal Reserve Bank of Chicago, working
paper, forthcoming.
Carron, A. S., 1982, “Financial crises: Recent expe
rience in U.S. and international markets,” Brookings
Papers on Economic Activity, Vol. 1982, No. 2,
pp. 395-418.
Hakkio, C. S., and W. R. Keeton, 2009, “Financial
stress: What is it, how can it be measured, and why
does it matter?,” Economic Review, Federal Reserve
Bank of Kansas City, Second Quarter, pp. 5-50.
Harvey, A., 1989, Forecasting, Structural Time Series
Models and the Kalman Filter, Cambridge, UK:
Cambridge University Press.
Hatzius, J., P. Hooper, F. Mishkin, K. L. Schoenholtz,
and M. W. Watson, 2010, “Financial conditions
indexes: A fresh look after the financial crisis,”
University of Chicago Booth School of Business,
Initiative on Global Markets, report, April 13, avail
able at http://research.chicagobooth.edu/igm/events/
docs/201 Ousmpfreport.pdf.
Illing, M., and Y. Liu, 2006, “Measuring financial
stress in a developed country: An application to Canada,”
Journal ofFinancial Stability, Vol. 2, No. 3, October,
pp. 243-265.
Laeven, L., and F. Valencia, 2008, “Systemic banking
crises: Anew database,” International Monetary Fund,
working paper, No. WP/08/224, November.
Minsky, H.P., 1986, Stabilizing an Unstable Economy,
New Haven, CT: Yale University Press.
Doz, C., D. Giannone, and L. Reichlin, 2006, “A
quasi maximum likelihood approach for large ap
proximate dynamic factor models,” European Central
Bank, working paper, No. 674, September.
Nelson, W. R., and R. Perli, 2007, “Selected indicators
of financial stability,” in Risk Measurement and Systemic
Risk, Frankfurt am Main, Germany: European Central
Bank, pp. 343-372.
Durbin, J., and S. J. Koopman, 2001, Time Series
Analysis by State Space Methods, Oxford, UK, and
New York: Oxford University Press.
Reinhart, C. M., and K. S. Rogoff, 2008, “This time
is different: A panoramic view of eight centuries of
financial crises,” National Bureau of Economic Research,
working paper, No. 13882, March.
El-Gamal, M. A., and A. M. Jaffe, 2008, “Energy,
financial contagion, and the dollar,” Rice University,
James A. Baker III Institute for Public Policy, work
ing paper, May.
Federal Deposit Insurance Corporation, 1997,
History’ of the Eighties—Lessonsfor the Future,
2 vols., Washington, DC.
__________ , 1984, Federal Deposit Insurance
Corporation: The First Fifty Years—A History’ of
the FDIC, 1933-1983, Washington, DC.
42
Schreft, S. L., 1990, “Credit controls: 1980,” Economic
Review, Federal Reserve Bank of Richmond, November/
December, pp. 25-55.
Shumway, R. H., and D. S. Stoffer, 1982, “An ap
proach to time series smoothing and forecasting using
the EM algorithm,” Journal of Time Series Analysis,
Vol. 3, No. 4, July, pp. 253-264.
Spero, J. E., 1999, The Failure of the Franklin
National Bank: Challenge to the International Banking
System, Washington, DC: Beard Books.
1Q/2011, Economic Perspectives
Stock, J. H., and M. W. Watson, 2002, “Forecasting
using principal components from a large number of
predictors,” Journal of the American Statistical
Association, Vol. 97, No. 460, December, pp. 1167-1179.
__________ , 1999, “Forecasting inflation,” Journal
ofMonetary’ Economics, Vol. 44, No. 2, October,
pp. 293-335.
Federal Reserve Bank of Chicago
Theil, H., 1971, Principles ofEconometrics, New York:
John Wiley and Sons.
Watson, M. W., and R. F. Engle, 1983, “Alternative
algorithms for the estimation of dynamic factor, mimic
and varying coefficient regression models,” Journal
ofEconometrics, Vol. 23, No. 3, December, pp. 385^100.
43
Understanding the Great Trade Collapse of 2008-09
and the subsequent trade recovery
Meredith A. Crowley and Xi Luo
Introduction and summary
In April 2009, the world economy appeared to be in
a free fall. Global trade in goods and services had
fallen 15.8 percent over the final two quarters of 2008
and the first quarter of 2009.1 This world trade collapse
had been the largest three-quarter decline of the past
40 years. Five months earlier, in November 2008, leaders
of the Group of Twenty (G-20)—20 large economies
that make up roughly 85 percent of the world’s economic
activity2—had met in Washington, DC, and pledged
to stabilize the world financial system and improve
coordination of macroeconomic responses to the global
financial crisis.3
Despite monetary easing and fiscal stimulus in
many economies, real economic activity continued
to deteriorate over the next few months. Reconvening
in April 2009 in London, the G-20 leaders had a full
agenda, which included the following topics: the role
of fiscal stimulus to promote recovery; the reform of
banking and financial regulation; and the strengthen
ing of the International Monetary Fund and the multi
lateral development banks (MDBs), such as the World
Bank Group. In addition, even though the world was
in the midst of an unprecedented globalfinancial crisis,
the problem of international trade was unusually
prominent on the agenda.
Among the commitments made by the G-20 in
London, two directly addressed international trade.
First, leaders promised to “ensure availability of at
least $250 billion over the next two years to support
trade finance through our export credit and investment
agencies and through the MDBs.”4 Second, they reaf
firmed a commitment made at the earlier Washington,
DC, summit to refrain from raising new barriers to trade
in goods and services. Finally, as part of the general
strategy to restore economic growth, they pointed out
that “an unprecedented and concerted fiscal expansion”
among the member economies would total $5 trillion
44
by the end of 2010.5 If declining trade simply reflected
declining economic activity, this fiscal expansion
would be expected to have an important impact on
global trade.
Previous work6 has documented what many
economists now refer to as the Great Trade Collapse
of 2008-09, and has analyzed its potential causes.
In this article, we review not only the unprecedented
collapse of world trade in 2008-09, but also the equally
dramatic trade recovery that took place in 2009-10.
We look at these events in a historical context, by
comparing them to previous trade contractions and
recoveries. To gain a better understanding of the links
between trade and broader economic conditions, we
look at changes in the trade-to-gross-domestic-product
(GDP) ratios of major economies across the globe be
fore, during, and after the Great Trade Collapse. Then,
we discuss three primary hypotheses that explain the
trade collapse: 1) a decline in aggregate demand for
all goods; 2) difficulties in obtaining trade finance; and
3) rising trade barriers. We consider how three distinct
policy actions—fiscal stimulus, funding for trade finance,
and a commitment to refrain from trade barriers—might
have affected both the collapse and the subsequent re
covery. Finally, we review four prominent examples
from the large literature examining the contributing
factors to the recent collapse of global trade.
Determining the relative degree to which the var
ious demand- and supply-side factors contributed to the
Great Trade Collapse is important for formulating the
optimal policy response. Economists would like to deter
mine if there are market failures or counterproductive
Meredith A. Crowley is a senior economist and Xi Luo is an
associate economist in the Economic Research Department
at the Federal Reserve Bank of Chicago. The authors thank
Gadi Barlevy, Lisa Barrow, Sam Kortum, and Ezra Oberfield
for thoughtful comments and suggestions.
2Q/2011, Economic Perspectives
policies specific to trade that the government can or
should correct. If research finds that weak domestic
demand (resulting from falling consumer income,
stronger preferences for saving over consumption, or
high unemployment) is the prime cause of the sharp
fall in trade, then there is not a clear mandate for gov
ernment intervention except, perhaps, actions to address
the overall recession. In contrast, if research shows
that trade finance problems are slowing down world
trade, the appropriate policy response might be inter
ventions by the government or nongovernmental or
ganizations in certain financial or insurance markets.
For example, governments could subsidize the price
of payment instruments, export credit insurance, or
even working capital loans. Finally, if analysis shows
that the government’s tariffs on imports or nontariff
barriers to trade are behind a sharp decline in trade,
then the best policy solution would be the removal
of these government interventions from international
goods markets.
According to the literature, the global collapse
in economic activity explains between 35 percent and
80 percent of the Great Trade Collapse. The analysis
we perform in this article estimates that declining
aggregate demand explains 35-50 percent of the
Great Trade Collapse. With regard to the recovery,
our analysis finds a quantitatively larger puzzle; rising
aggregate demand explains only 25-40 percent of the
recovery in imports. The findings of the literature on
the role of trade finance in the collapse are mixed, with
one paper finding that tighter financial conditions likely
had a moderate negative effect on trade volumes during
the financial crisis of 2008-09. Further, in this article,
we document the evolution of antidumping trade restric
tions imposed by the United States and Canada over
the past 40 years and conclude that there was no sig
nificant increase in border restrictions by these two
countries in 2008 or 2009. Thus, trade protection by
these countries was not a cause of the collapse. In terms
of the dramatic recovery in trade, the absence of explicit
border barriers at least allowed the recovery to progress
unhindered. The conclusion that changing aggregate
demand was the major cause of both the dramatic col
lapse in trade volumes in 2008-09 and the spectacular
recovery in 2009-10 suggests that of all the policy
actions, fiscal stimulus likely had the largest impact
on the trade recovery.
What was the Great Trade Collapse?
In this section, we document some stylized facts
about the Great Trade Collapse of 2008-09 and the sub
sequent recovery. Panel A of figure 1 documents the
timing and magnitude of the Great Trade Collapse.
Federal Reserve Bank of Chicago
The plotted series is the seasonally adjusted quarterly
level of world trade measured in trillions of 2005 U.S.
dollars. World trade of goods and services is defined
as (A + A/)/2, where X is world exports of goods and
services and M is world imports of goods and services.
The V-shaped path toward the end of panel A corre
sponds to the collapse in world trade during the period
2008:Q2-2009:Q2 and the equally rapid recovery from
2009:Q2 onward. This world trade series from the
Organisation for Economic Co-operation and Develop
ment (OECD), which starts in 1968:Q2, demonstrates
a clear upward trend. The level of world trade in
2010:Q3 is more than 15 times the level in 1968:Q2.
While international trade has been trending upward
for more than four decades, with an annual growth
rate of 6.48 percent, episodes of contraction have not
been uncommon. Between 1974:Q2 and 1975:Q2, the
world trade level declined by 7.65 percent; between
198O:Q1 and 1980:Q3, it slid by 3.34 percent; between
1981:Q4 and 1982:Q4, it slipped by 3.12 percent;
and between 2000 :Q4 and 2001:Q4, it decreased by
3.51 percent. The Great Trade Collapse, which occurred
between 2008:Q2 and 2009:Q2, was more severe than
all the previous tumbles—the volume of world trade
plummeted by 17.20 percent from peak to trough.
In panel B of figure 1, the log of world trade in
trillions of 2005 U.S. dollars is plotted. This series
displays a clear linear trend. Notice that during the
2000s, trade growth stood above the trend line until
the collapse of 2008-09. Although a rapid recovery
began after 2009:Q2, world trade has yet to return to
its long-run linear trend.
Next, we turn to the United States. Panel A of
figure 2 shows real seasonally adjusted U.S. imports
and exports. Like the rest of the world, the United
States has seen fast growth in trade over the past few
decades. From 1965 through 2010, U.S. imports grew
at an annual rate of 6.03 percent and U.S. exports grew
at an annual rate of 5.92 percent. During the Great
Trade Collapse (2008:Q2-2009:Q2), U.S. real imports
declined by 18.3 percent while U.S. real exports dropped
by 14.7 percent. Given the rapid growth in trade over
the previous five decades, the magnitude of the col
lapse in exports and imports was truly astonishing.
Panel B of figure 2 shows the log levels of U.S.
real imports and exports, which both display linear
upward trends over time. Notice that the bumps and
wiggles in the series for the United States are more
apparent than in their counterparts for world trade in
panel B of figure 1. These differences between world
trade and U.S. trade measures are due to the fact that
in world trade flows, a decline in one country’s trade
volume is often offset by growth in another’s.
45
FIGURE 1
World trade, 1968-2010
A. World trade
B. Log of world trade
trillions of 2005 U.S. dollars, seasonally adjusted
Notes: World trade is the sum of world exports in goods and services and world imports in goods and services divided by two. In each panel,
the two dashed vertical lines indicate the peak and trough of the Great Trade Collapse (2008:Q2-2009:Q2). In panel B, the straight black line
indicates the long-run linear trend.
Source: Authors’ calculations based on data from the Organisation for Economic Co-operation and Development, Main Economic Indicators,
from Haver Analytics.
FIGURE 2
U.S. trade, 1965-2010
B. Log of U.S. trade
1970
’80
’90
2000
-------
'10
Exports
-------
Imports
Note: The shaded areas indicate official U.S. periods of recession as identified by the National Bureau of Economic Research.
Source: Authors’ calculations based on data from the U.S. Bureau of Economic Analysis, National Income and Product Accounts
of the United States, from Haver Analytics.
46
2Q/2011, Economic Perspectives
FIGURE 3
Trade contractions and recoveries
B. World
A. United States
normalized value of U.S. trade
normalized value of world trade
quarters since trough
quarters since trough
-6- 1975:Q2
-©-
1991:Q1
-©- 1980:Q3
-6-
2001 :Q4
-©- 1982:Q4
-©-
2009:Q2
Notes: Episodes of trade contraction and recovery for both the United States and the world are indicated by their trough dates. Panel A is
based on U.S. trade data in figure 2. Panel B is based on world trade data in figure 1. For each panel’s vertical axis, the data are normalized
to be equal to 100 for the indicated year and quarter.
Sources: Authors’ calculations based on data from the U.S. Bureau of Economic Analysis, National Income and Product Accounts of the
United States, and Organisation for Economic Co-operation and Development, Main Economic Indicators, from Haver Analytics.
Trade contractions and recoveries
in historical perspective
How does the most recent trade collapse compare
with previous episodes of trade contraction? And how
does the current recovery in trade compare with previous
recoveries? In figure 3, we present “spider graphs” that
allow us to compare the magnitude and speed of differ
ent trade contractions and recoveries. Panel A of figure 3
presents several U.S. trade contractions and recoveries,
while panel B of figure 3 presents trade contractions
and recoveries for the world. In both panels A and B
of figure 3, we normalize real, seasonally adjusted,
quarterly data on trade, defined as (X + M)I2, to be equal
to 100 in the quarter identified as the trough of each
U.S. trade contraction. We identified the following quar
ters as the troughs of U.S. episodes of trade contraction
and recovery: 1975:Q2, 1980:Q3, 1982:Q4, 1991 :Q1,
2001:Q4, and 2009:Q2. Next, we investigate what
happened to the volume of U.S. and world trade four
quarters before and five quarters after these identified
nadirs for U.S. trade. Numbers on the horizontal axes
represent the number of quarters before and after the
trough date; therefore, the number zero corresponds
to the troughs. Numbers to the left of zero generally
Federal Reserve Bank of Chicago
correspond to a period of decline in trade volume. Anal
ogously, numbers to the right of zero generally corre
spond to a period of recovery in trade volume. We
refer to each episode of trade contraction and recovery
by its trough date.
In both panels, the solid black line stands out.
The black lines (representing the 2009 :Q2 episode)
depict the changes in trade volume during the Great
Trade Collapse of 2008-09 (and the subsequent re
covery) for the United States and the world in panels
A and B, respectively. A closer look at these spider
graphs reveals the following facts.
First, for both the United States and the world,
the recent trade collapse is the most severe decline in
trade since the late 1960s, in terms of both magnitude
and speed. Notice that for both the United States and
the world, sustained trade declines do not last more than
four quarters. For the United States, the 1975:Q2,
1982:Q4, and 2001:Q4 episodes all have four quar
ters of contraction. In contrast, for the 1980:Q3 and
1991 :Q 1 episodes in the United States, contractions
lasted for only two quarters. The patterns of contraction
in world trade are almost identical to those in the U.S.
trade. An exception is the 1991 :Q 1 episode in which
world trade never experienced a decline.
47
Second, despite its huge magnitude,
TABLE 1
the Great Trade Collapse does not stand
Averages of annualized quarterly growth rates
out as more protracted than previous epi
of trade
sodes. Thus, a greater amount of trade de
Episode
struction occurred in a period of typical
by trough
__ United States_____
_______ World________
duration for trade decline. One way to see
date
Contraction
Recovery
Contraction
Recovery
this point is to compare the averages of the
(..................................... percent....................................... )
annualized quarterly growth rates of trade
1975:Q2
-12.3
15.3
-6.9
13.4
during the four quarters before the identi
1980:Q3
-0.4
4.3
1.1
6.3
fied trough dates (see table 1). During
1982:Q4
-7.8
15.7
-2.9
9.8
the Great Trade Collapse (the 2009:Q2
1991 :Q1
-0.6
9.0
4.7
7.7
episode in table 1), U.S. trade fell, at
2001 :Q4
-9.5
7.4
-3.4
8.4
an average annualized quarterly rate of
2009:Q2
-15.8
16.0
-13.6
14.4
-15.8 percent, and world trade dropped,
Notes: Trade is (X + M)/2, where X is exports of goods and services and M
at an average annualized quarterly rate of
is imports of goods and services. The underlying U.S. data series is reported
in billions of 2005 chained U.S. dollars, seasonally adjusted. The underlying
-13.6 percent. The trade contraction fol
world data are reported in billions of 2005 U.S. dollars, seasonally adjusted.
lowing in the wake of the oil crisis of 1973
The averages of the annualized quarterly growth rates of trade are calculated
during each trade episode’s contraction (four quarters before the trough) and
(the 1975:Q2 episode in table 1), the most
recovery (five quarters after the trough). For the world’s 2009:Q2 episode,
similar in terms of magnitude, saw U.S.
the recovery rate is calculated for four quarters after the trough.
Sources: Authors’ calculations based on data from the U.S. Bureau of Economic
trade fall, at an average annualized rate
Analysis, National Income and Product Accounts of the United States, and
of-12.3 percent.
Organisation for Economic Co-operation and Development, Main Economic
Indicators, from Haver Analytics.
Third, let us take a look at the right
hand side of each panel in figure 3 and
examine the recovery that followed each
collapse. We notice that following the
Figure 4 disentangles the U.S. episodes of trade con
nadir of the Great Trade Collapse (2009 :Q2), despite
traction and recovery into spider graphs of imports
a remarkably fast recovery rate, as of2010:Q2, both
(panel A) and exports (panel B). All import episodes
U.S. and world trade have yet to return to their pre
have a V-shaped path, while not all export episodes
collapse levels. For world trade, in all previous con
display this pattern. Apparently, imports played the
tractions, trade volumes rebounded to their pre-collapse
more significant role in shaping the U.S. trade con
levels within four quarters.
traction episodes displayed in figure 3.
For the United States, a slow recovery in trade
Let us first focus on U.S. imports in panel A of
is not unprecedented. After the trade contraction
figure 4. Interestingly, with respect to imports, the Great
associated with the dot-com recession of 2001 (that
Trade Collapse (the 2009 :Q2 episode) looks similar
is, the 2001 :Q4 episode in figure 3), it took eight quar
to the trade contraction associated with the oil shock
ters for trade to rebound to its pre-contraction level.
of 1973 (the 1975:Q2 episode). The magnitudes of
The trade recovery following the Great Trade Collapse
the contractions over the four quarters before the trough
has been faster than that following the dot-com bust.
date are similar. In fact, the average of the annualized
Five quarters after the nadir in 2009:Q2, U.S. trade
quarterly growth rates of U.S. imports was -18.4 percent
volume had returned to 99.3 percent of its 2008:Q2
during the 1975:Q2 episode versus -17.2 percent during
level. Given the severity of the decline, this fivethe 2009 :Q2 episode. The magnitudes of the rebounds
quarter rally has been impressive.
over the five quarters after the trough date are not too
Finally, figure 3 suggests that there may be a syn
far off from each other. The average of the annualized
chronicity between U.S. and world trade. The U.S.
quarterly growth rates of U.S. imports was 24.9 percent
trough dates are identical with the world trough dates
for the 1975:Q2 episode versus 17.8 percent for the
on most occasions. However, it is not clear from this
2009:Q2 episode. Still, compared with the previous epi
figure if this synchronicity is due to the United States’
sodes of trade contraction, the collapse in U.S. imports
large share of world trade or due to changes in foreign
in 2008-09 was among the most severe. When exam
trade flows that are truly synchronous with U.S. trade
ining the rebounds in imports of the various episodes,
flows. We return to this issue later.
we see that imports grew firmly, but not stunningly, after
We now shift gears to examine U.S. imports and
the Great Trade Collapse. On the one hand, a recovery
exports in order to understand the Great Trade Collapse
of 17.8 percent for the 2009:Q2 episode has been much
and the subsequent recovery from another angle.
48
2Q/2011, Economic Perspectives
FIGURE 4
U.S. trade contractions and recoveries: Imports and exports
B. Exports
normalized value of exports
A. Imports
normalized value of imports
quarters since trough
quarters since trough
-©- 1975:Q2
-9-
1991:Q1
-6- 1980:Q3
-©-
2001 :Q4
-©- 1982:Q4
—Q—
2009:Q2
Notes: Episodes of trade contraction and recovery for the United States are indicated by their trough dates, as in figure 3. Panel A is
based on the import volume series and panel B is based on the export volume series in figure 2. For each panel’s vertical axis, the
data are normalized to be equal to 100 for the indicated year and quarter.
Source: Authors’ calculations based on data from the U.S. Bureau of Economic Analysis, National Income and Product Accounts
of the United States, from Haver Analytics.
faster than those of the 1980:Q3, 1991:Q1, and
2001:Q4 episodes; on the other hand, the recovery
speed for the 2009 :Q2 episode has not been as fast as
those for the 1975;Q2 and 1982;Q4 episodes.
Next we turn to the export side in panel B of
figure 4. Note that the Great Trade Collapse and the
recovery following it (the 2009:Q2 episode) had the
steepest and most symmetric V-shaped path around
the trough date relative to all previous episodes. This
makes the Great Trade Collapse and subsequent recovery
look unique. Over the four quarters before the 2009:Q2
trough date, the average of the annualized quarterly
growth rates of exports was -13.9 percent, the largest
rate of decline seen over the past four decades. The
recovery over the five quarters after the 2009:Q2 trough
date has been fast, with an average of the annualized
quarterly growth rates of 12.7 percent. The momentum
of the export recovery was rapid in the beginning but
gradually faded. The export series during the Great
Trade Collapse and subsequent recovery features a
quick collapse and a quick rebound.
The export contractions in the 2001 :Q4 and 1982:Q4
episodes look similar to that of the 2009 :Q2 episode,
although both of the earlier episodes feature slow
Federal Reserve Bank of Chicago
recoveries. In contrast, the 1975;Q2, 1980:Q3, and
1991 :Q 1 episodes do not have V-shaped paths. Take
the 1975;Q2 episode, for example. During the collapse
period, exports slid for a quarter, rebounded for two
consecutive quarters, and then declined for two more
quarters (past the trough date of imports for that epi
sode). One quarter into the recovery, a brief reversal set
in before a two-quarter rally that finally brought the
export volume back to the level of 1974:Q2. Exports
in the 1980:Q3 and 1991:Q1 episodes experienced little
or no decline. Therefore, the brief trade contractions
in the 1980:Q3 and 1991 :Q1 episodes can be attributed
almost exclusively to contractions in imports.
To summarize, the behavior for U.S. imports dur
ing the Great Trade Collapse and the subsequent recov
ery look similar to that of previous episodes. However,
the V-shaped pattern of U.S. exports during the Great
Trade Collapse and the subsequent recovery bears little
resemblance to the behavior of exports in previous epi
sodes. The unique path of exports during the 2009:Q2
episode appears to be driven by the strength of the
2008-09 global recession, which we explore in more
detail in the next section.
49
Changes in trade and GDP
Trade volume usually rises or falls in accordance
with the direction of the general economy, so we want
to examine this interaction. For U.S. trade levels, if
we refer to figure 2 (on p. 46), for example, we see
that trade contractions usually occur during recessions.
How do we think of a trade contraction in the context
of broader economic conditions? For any country, by
summing up imports and exports and then dividing
this quantity by GDP, we obtain that country’s trade-toGDP ratio. Multiplying by 100 allows us to express
this ratio as a percent. Figure 5 shows the nominal
trade-to-GDP ratios of the United States, France,
Japan, and Germany over the past few decades.
Let us focus on the U.S. experience plotted in
panel A of figure 5. This ratio was 8.85 percent in
1965:Q1 and peaked in2008:Q3 at 31.88 percent.
The upward trend in the evolution of this ratio indicates
that the growth in trade volume has outpaced the growth
in GDP over the past few decades; trade’s role in the
broader economy has expanded steadily. The trade-toGDP ratio can be thought of as a measure of the open
ness of an economy to trade. The fact that the trade-toGDP ratios for the United States, France, Japan, and
Germany have all been trending upward over time shows
that these countries have become more and more open
to trade as part of their economic activities. This rise
in openness is often referred to as globalization.
Each country’s path to globalization is subject
to its own historical idiosyncrasies. For example,
the declines in the United States’ trade-to-GDP ratio
occur close to U.S. recessions. During the period
1974:Q4—1975:Q3, around the time of the first oil crisis,
the trade-to-GDP ratio decreased from 17.6 percent to
15.4 percent. Around the time that the dot-com bubble
burst, in the period 2000:Q3-2001:Q4, the trade-toGDP ratio decreased from 26.3 percent to 22.0 percent.
Finally, around the time of the global financial crisis,
during the period 2008:Q3-2009:Q2, the trade-to-GDP
ratio plummeted from 31.9 percent to 24.1 percent.
For France (figure 5, panel B), fluctuations in
the trade-to-GDP ratio follow a similar pattern to that
observed for the United States. Starting at 26.3 percent
in 1965:Q1, France’s trade-to-GDP ratio increased
steadily over time, reaching 43.7 percent in 1974:Q3.
When the oil shock set in, the trade-to-GDP ratio slid
to 35.9 percent in 1975:Q3, and it did not surpass the
pre-collapse level until 1980:Q 1—five and a quarter
years after the trough. For France, whenever there is
a drop in the trade-to-GDP ratio, it takes a relatively
long time to recover. France experienced a plodding
recovery from the trade contraction of the early 2000s.
In 2008:Q3, France’s trade-to-GDP ratio stood at
50
56.6 percent, but it was crushed to 47.3 percent within
three quarters. For France, the Great Trade Collapse
appears to have precipitated a dip in the trade-to-GDP
ratio following a relatively weak recovery from the
earlier decline that coincided with the United States’
dot-com recession.
Turning to Japan (figure 5, panel C), we see that
the nominal trade-to-GDP ratio started from almost
30 percent in the early 1980s. This ratio dropped dras
tically following the 1985 Plaza Accord, under which
the Japanese yen started to appreciate against other
major world currencies. Japan’s trade-to-GDP ratio
dropped from 27.3 percent in 1984:Q4 to 16.8 percent
in 1988:Q1. After rising for a few years, this ratio
took another dip in the early 1990s, when it declined
to a low of 15.6 percent in 1993:Q4. Following that
dip, the trade-to-GDP ratio recovered steadily. Since
2001:Q4, Japan’s trade-to-GDP ratio had risen quickly,
to a peak in 2008:Q3 of 38.6 percent. During the Great
Trade Collapse, the trade-to-GDP ratio took a nose dive.
Four quarters after the trough in 2009:Q2, Japan’s
trade-to-GDP ratio had recovered only about half of
the lost ground, standing at 30.0 percent.
Germany’s trade-to-GDP ratio (figure 5, panel D)
has trended upward, starting from 39.5 percent in
1968:Q1 to reach a peak of 90.9 percent in 2008:Q3.
The reunification of Germany in the early 1990s
knocked this ratio down from 63.1 percent in 1990:Q4
to 44.3 percent in 1993:Q4. Since then, the openness
of Germany’s economy to trade had increased signifi
cantly until the Great Trade Collapse. After peaking in
2008:Q3, Germany’s trade-to-GDP ratio fell to 74.6 per
cent in 2009 :Q2, before beginning a sharp recovery.
Examining the experiences of four major world
economies displayed in figure 5, we conclude that in
ternational trade has become more and more important
to the global economy over time. What caused inter
national trade to grow so explosively? In the post-World
War II era, several factors have facilitated this meteoric
growth in international trade: 1) the decline in tariffs
under the General Agreement on Tariffs and Trade/
World Trade Organization (GATT/WTO) system7
(Crowley, 2003; and Subramanian and Wei, 2007),
as well as a number of preferential trade agreements;
2) the decline in transportation costs (Hummels, 2001,
2007; and Levinson, 2006); 3) the rise of vertical spe
cialization8 facilitated by the first two factors (Yi, 2003);
and 4) the decline in communication costs (Freund
and Weinhold, 2000).
Given the rising openness to trade around the world
depicted in figure 5, the Great Trade Collapse stands out
not only because of its magnitude, but also because it ap
pears to have been highly synchronized across countries.
2Q/2011, Economic Perspectives
FIGURE 5
The ratio of trade to gross domestic product (GDP) for selected countries
B. France
percent
C. Japan
percent
D. Germany
percent
Notes: In each panel, the trade-to-GDP ratio is defined as the sum of nominal imports and nominal exports divided by nominal GDP.
Also, in each panel, the shaded areas indicate official U.S. periods of recession as identified by the National Bureau of Economic Research.
Sources: Authors’ calculations based on data from the U.S. Bureau of Economic Analysis, Institut National de la Statistique et des Etudes
Economiques of France, Cabinet Office of Japan, and Deutsche Bundesbank of Germany, from Haver Analytics.
Let us now examine the synchronicity of the
Great Trade Collapse and the subsequent recovery by
reviewing the experience of a broader range of coun
tries. Figure 6 is a scatter plot of the percentage change
in trade versus the percentage change in real GDP over
the period 2008:Q2-2009:Q2 for 29 countries.9 Three
important facts emerge from this picture.
First, the decline in trade was broadly spread
across this entire set of countries. During this period,
the least affected country plotted, that is, Brazil, had
a change in trade of more than -7.5 percent. The most
affected country, that is, Mexico, had a change in trade
of-26.1 percent. The United States’ trade collapse,
Federal Reserve Bank of Chicago
amounting to a change of-15.0 percent, fell right in
the midrange of this cross section of countries.
Second, with the exception of Australia, Poland,
India, and Brazil, all countries displayed here experi
enced declines in their GDP as well. Mexico again led
the group, with a change of-10.0 percent. The United
States experienced a -4.1 percent change in its GDP.
Among the larger economies, Japan experienced a
-5.9 percent change in GDP (as well as a -25.0 percent
change in trade).
Third, a fitted line through this scatter plot has a
slope of 0.96, which indicates that a 1 percent decline
in GDP is associated with a 0.96 percent decline in
51
FIGURE 6
The change in real trade vs. the change in real GDP during the Great Trade Collapse
-10
-5
0
percentage change in real GDP
AUS—Australia
DEN—Denmark
GRE—Greece
KOR—South Korea
AUT—Austria
BRA—Brazil
ESP—Spain
FIN—Finland
HUN—Hungary
IND—India
MEX—Mexico
NED—Netherlands
BEL—Belgium
CAN—Canada
FRA—France
GBR—United Kingdom
ISL—Iceland
ITA—Italy
NZL—New Zealand
NOR—Norway
CZE—Czech Republic
GER—Germany
JPN—Japan
POL—Poland
POR—Portugal
SVK—Slovakia
SUI—Switzerland
TUR—Turkey
USA—United States
of America
Notes: Both the changes in real trade and real gross domestic product (GDP) are measured over the period 2008:02-2009:02.
The dashed black line indicates the relationship between trade and GDP over this period; the shaded region indicates the 95 percent
confidence band around the regression line.
Source: Authors’ calculations based on data from the Organisation for Economic Co-operation and Development, Main Economic
Indicators, from Haver Analytics.
trade. This picture highlights the global synchronicity
of both the Great Recession and the Great Trade
Collapse.10
Figure 7 plots the recovery following the Great
Trade Collapse, using data from 2009 :Q2 through
2010:Ql. From this figure we see that most countries
were recovering from the Great Recession during this
period; only Spain, Greece, and Israel saw GDP de
creasing over the period 2009:Q2-2010:Ql. India
(omitted from the figure) was a strong outlier, with
dramatic GDP growth of 21.8 percent over this period.
The figure also demonstrates that the recovery of trade
has been widespread and, for many countries, strong.
Only Greece and Finland continued to experience
declines in trade after 2009:Q2.
To conclude, the highly synchronized nature of the
global trade collapse that occurred in 2008-09 and the
subsequent recovery suggests that analytical models
of the Great Trade Collapse should be global in nature.
52
What caused the Great Trade Collapse
and the subsequent recovery?
What was behind the sharp decline in world
trade that began in the second quarter of 2008? And
what is behind the amazingly quick recovery in trade
that we are experiencing today? The facts that we have
gleaned from the data can help guide our analysis. First,
we know that the Great Trade Collapse was extremely
severe and steep by historical standards. Second, trade
fell more dramatically than GDP around the world.
Third, compared with previous episodes in which U.S.
imports and exports fell, this trade collapse was much
more highly synchronized around the world. In forming
hypotheses to explain the causes of a phenomenon
like the Great Trade Collapse, economists often begin
with a simple supply-and-demand framework. If the
quantity of imports falls during a recession, one likely
culprit for this decrease is the decline in consumers’
incomes, which reduces consumer demand for all goods,
2Q/2011, Economic Perspectives
FIGURE 7
The change in real trade vs. the change in real GDP after the Great Trade Collapse
percentage change in real GDP
Notes: For the legend explaining the country abbreviations, see figure 6. India (IND) is not featured here because it is an outlier.
Data for Austria (AUT), Norway (NOR), and Portugal (POR) were not available. Both the changes in real trade and real gross
domestic product (GDP) are measured over the period 2009:02-2010:Q1. The dashed black line indicates the relationship be
tween trade and GDP over this period; the shaded region indicates the 95 percent confidence band around the regression line.
Source: Authors’ calculations based on data from the Organisation for Economic Co-operation and Development, Main Economic
Indicators, from Haver Analytics.
including imports. In other words, as consumers tight
ened their belts and bought fewer domestically produced
goods, they also chose to buy fewer imported goods.
However, we know that during the Great Trade Collapse,
imports fell much more rapidly than income.
Are there complicating factors behind a decline
in consumer demand for imports? Possibly. As can be
seen in figure 1 (p. 46), global trade began to take off
in the mid-1990s. While there were many forces at work,
a key element in this transition was the rise of global
supply chains. Because companies now spread their
production processes across multiple countries, the
production of a specific good—for example, a car—
involves multiple border crossings of a partially com
pleted car that becomes more valuable with every step
in the production process and every border crossing.
Because customs agencies record the total value of
every object that crosses the border and not the value
added to the object during its most recent trip to a coun
try, the value of trade recorded by national customs
agencies has grown more rapidly than GDP as more and
more companies and industries have spread their pro
duction processes across many countries. It is difficult
Federal Reserve Bank of Chicago
to precisely measure the importance of trade in inter
mediate goods (for example, an engine or brake for a
car). That said, one OECD study11 estimates that the
average annual growth rate of trade in intermediate
goods among OECD members was 6.20 percent over
the period 1995-2006, whereas the average annual
growth rate in the trade of final consumption goods was
only 5.87 percent over the same period. This finding
suggests that the share of intermediate goods in total
trade flows has been increasing as global supply chains
have spread.
In the Great Trade Collapse, we might have been
observing the rapid unwinding of these global supply
chains. Within a vertically integrated international
economy,12 a simple fall in consumer demand for im
ports would have been magnified through the global
supply chains. For every car that is not produced and
sold to a consumer, trade flows as measured by customs
authorities fell by more than the final value of the car
because that car, which would have crossed several
borders during its production, did not cross any borders.
So, in addition to falling consumer demand, this com
plication generated by various multicountry production
53
processes may have played a significant role in pre
cipitating and/or exacerbating the Great Trade Collapse.
Another complicating feature of the demand side
is that there are compositional differences between
imports and national income, or GDP. Consider the
United States’ imports and national income. The vast
majority of imports into the United States are goods—
for example, food, clothing, cars, and electronics—
but some of these imports are services—for example,
education, travel, and business consulting services.
Our national income consists largely of services—for
example, health care and education—with goods playing
a much smaller role in our economy today than they
have in the past. We might expect that consumption
of some domestically produced services like health care
is more recession-proof than the consumption of typi
cally imported goods like televisions and refrigerators.
How much of the Great Trade Collapse (and the sub
sequent recovery) was due to a difference in the rela
tive composition of tradable versus domestically
produced goods and services?
Returning to our simple framework, we note that
the other likely cause of the recent trade collapse would
be some type of disruption on the supply side—that is,
some factor that affects the firms that are producing
goods and shipping them to consumers and retail out
lets. During the recent global recession, which started
with a global financial crisis, the costs associated with
exporting were carefully monitored for their potential
impact on trade flows. Because the crisis was a finan
cial one, governments and international organizations,
such as the WTO and World Bank, tried to collect in
formation on the costs of financing trade. Given the
tight financial environment during the crisis, did firms
face difficulty in obtaining different types of financing
for their international shipments? In addition, were
there problems associated with rising trade protection
during the recent recession? It is widely known that
the United States increased import tariffs during the
Great Depression and that this likely worsened the
severity of the depression during the 1930s. Did
something similar happen this time around to cause
or exacerbate the Great Trade Collapse?
Demand-side explanations
Is it surprising that trade collapsed during the
recent global recession? As we discussed previously,
the Great Trade Collapse was coincident with the largest
decline in world GDP in decades. Should we not have
expected that consumers, who buy less of everything
during a recession, would also buy fewer imported
goods? How can economists assess this problem on
the demand side quantitatively?
54
To predict how exports or imports will change
in the future, economists routinely estimate trade
elasticities. Trade elasticities with respect to income
measure how much a country’s imports or exports
will change in response to changes in national income.13
For example, the import elasticity with respect to in
come is a number that specifies how much imports
will increase in response to a 1 percent increase in the
total income of a country. Economic theory posits that
this elasticity is positive. That is, an increase in a coun
try’s income leads it to buy more from foreign countries.
Moreover, an income elasticity of imports that is equal
to one implies that imports increase proportionately
with national income.
For the past several decades, estimates of the im
port elasticity with respect to income for the United
States have ranged from 1.5 to slightly more than two.14
That is, in the United States, imports respond more
than proportionately to changes in income. Precisely
how much more depends on the exact value of the
elasticity. In table 2, we list reported estimates of the
import elasticity with respect to income for the United
States by several different researchers. Using infor
mation on the decline in U.S. GDP over the period
2008:Q2-2009:Q2, we can predict how large the U.S.
decline in imports must have been in order to be in
line with historical norms. Specifically, the actual
cumulative change in U.S. GDP over this time period
was -4.1 percent. In table 2, we use import elasticities
with respect to income to predict the decline in imports
during the Great Trade Collapse (2008:Q2-2009:Q2).
Predictions for the change in U.S. imports range from
a low of-6.2 percent, using a historical estimate from
Houthakker and Magee (1969), to a high of-9.4 percent,
using the more recent estimate from Chinn (2004)
(the fourth column of table 2). But the cumulative
change in imports over the period 2008:Q2-2009:Q2
was actually -18.3 percent. Estimates of the import
elasticity with respect to income indicate that the decline
in U.S. national income during the Great Trade Collapse
can explain only about 35 percent to 50 percent of the
decline in U.S. imports (see the last column of table 2).
This simple analysis of demand-side factors tells us
that imports fell about twice as much as we would
have expected!
How unusual is the Great Trade Collapse in this
regard? That is, if we examine the other major contrac
tions in U.S. imports since the 1970s, how do they com
pare? Table 3 compares the Great Trade Collapse with
five previous import contraction episodes in the United
States. The first column lists the trough date of each
of the six major contractions in U.S. imports reported
earlier in figures 3 and 4 (pp. 47 and 49). The second
2Q/2011, Economic Perspectives
TABLE 2
Predicted vs. actual change in U.S. imports, 2008:Q2-2009:Q2
Previous research
Houthakker and Magee (1969)
Hooper, Johnson, and Marquez (2000)
Chinn (2004)
Cardarelli and Rebucci (2007)
Crane, Crowley, and Quayyum (2007)
Sample
period
Import
elasticity with
respect to income
Predicted percent
change in imports
Predicted change
in imports/actual
change in imports
1951-66
1961-94
1975-2003
1972-2006
1960-2006
1.51
1.79
2.29
2.03
1.93
-6.2
-7.3
-9.4
-8.3
-7.9
0.34
0.40
0.51
0.45
0.43
Note: See the text for further details.
Source: Authors’ calculations based on data from the U.S. Bureau of Economic Analysis, National Income and Product Accounts of the
United States, from Haver Analytics.
TABLE 3
Predicted vs. actual change in U.S. imports during episodes of trade contraction
Trough date of the import
contraction episode
1975:Q2
1980:Q3
1982:Q4
1991 :Q1
2001 :Q4
2009:Q2
Percent change
in U.S. gross
domestic product
Predicted
percent change
in imports
Actual
percent change
in imports
-1.8
-1.6
-1.4
-1.0
0.4
-4.1
-3.5
-3.1
-2.7
-1.9
0.8
-7.9
-19.5
-12.0
-3.9
-4.3
-7.8
-18.3
Predicted change
in imports/actual
change in imports
0.18
0.26
0.69
0.44
N.A.
0.43
Notes: N.A. indicates not applicable. See the text for further details.
Source: Authors’ calculations based on data from the U.S. Bureau of Economic Analysis, National Income and Product Accounts of the
United States, from Haver Analytics.
column presents the cumulative decline in U.S. real
GDP over the four quarters before the trough date.
The third column displays the implied change in U.S.
real imports over the four quarters before the trough
date; we derive these values by multiplying the change
in real GDP in the second column with the estimate
of the import elasticity with respect to income (1.93)
from Crane, Crowley, and Quayyum (2007). In the
fourth column, we report the actual percent change
in U.S. real imports over the four quarters before the
trough date. Interestingly, in almost all cases the actu
al changes in trade were substantially larger than the
predicted changes in the third column. The last column
lists the ratio of the predicted change in imports to the
actual change in imports, which reveals how much
of the decline in imports over the four quarters con
sidered may be due to a decline in GDP over the same
period. It appears that declining aggregate demand
varies considerably in its importance as a cause for
these trade declines. Thus, other factors, such as
changing relative prices, trade barriers, or costs of
Federal Reserve Bank of Chicago
conducting international trade, must also contribute
to these trade contractions.
An import elasticity analysis of the trade recov
ery from 2009:Q2 through 2010:Q2 leaves us with
a quantitatively even larger puzzle. Over the period
2009:Q2-2010:Q2, U.S. GDP grew 3.0 percent, but
U.S. imports of goods and services skyrocketed up
17.4 percent. Turning to table 4, we see that this dramatic
increase in imports cannot be well explained simply
by an improvement in aggregate demand. Table 4’s
final column indicates that the increase in U.S. national
income can explain only about one-quarter to 40 percent
of the recovery following the Great Trade Collapse.
As we stated before, during the Great Trade
Collapse, a decline in U.S. aggregate income can ex
plain only about half of the decline in imports. Similarly,
when U.S. GDP began to recover after this collapse,
U.S. imports surged well beyond the improvement
predicted by the United States’ import elasticity with
respect to income. So what other forces account for
55
TABLE 4
Predicted vs. actual change in U.S. imports, 2009:Q2-2010:Q2
Previous research
Houthakker and Magee (1969)
Hooper, Johnson, and Marquez (2000)
Chinn (2004)
Cardarelli and Rebucci (2007)
Crane, Crowley, and Quayyum (2007)
Sample
period
Import
elasticity with
respect to income
Predicted percent
change in imports
Predicted change
in imports/actual
change in imports
1951-66
1961-94
1975-2003
1972-2006
1960-2006
1.51
1.79
2.29
2.03
1.93
4.5
5.4
6.9
6.1
5.8
0.26
0.31
0.40
0.35
0.33
Note: See the text for further details.
Source: Authors’ calculations based on data from the U.S. Bureau of Economic Analysis, National Income and Product Accounts of the
United States, from Haver Analytics.
the unexplained movements in U.S. imports during
the collapse and the recovery?
To address this question, we next examine some
what detailed data on U.S. imports in order to identify
compositional changes in imports that occurred over
the period 2008-10.
First, let us examine changes in U.S. imports of
goods and services. Figure 8 plots real U.S. imports
of goods versus services since the mid-1960s. A sig
nificant decline and recovery occurred for imports of
goods over the period 2008-10. Although the services
industry plays a large role in the U.S. economy today,
goods cross borders more often than services and were
hit harder during the Great Trade Collapse. Over the
period 2008:Q2-2009:Q2, U.S. imports of goods fell
by 21.1 percent, while U.S. imports of services fell
by only 3.5 percent. On the recovery side, from
2009:Q2 through 2010:Q3, U.S. imports of goods in
creased 18.4 percent, while U.S. imports of services
increased by only 6.2 percent. From these observations,
we conclude that the Great Trade Collapse and the
subsequent recovery were driven by changes in the
trade of goods.
Naturally, oil is suspected as one large factor be
hind the Great Trade Collapse. Earlier we discussed
the oil crisis of 1973 as a factor in a previous large
trade contraction. How big a role did oil play in the
most recent trade episode? Figure 9 plots real U.S.
imports of petroleum and nonpetroleum goods in billions
of chained 2005 dollars. We see that, although petro
leum imports have grown over time, their importance
as a share of all imports has declined. In 1974:Q2, oil
represented 45.0 percent of U.S. goods imports, but by
the time of the Great Trade Collapse in 2008:Q2, oil’s
share of U.S. goods imports had fallen to 13.4 percent.
Moreover, the peak-to-trough decline in oil imports
of 13.4 percent during the trade contraction of
1974:Q2-1975:Q2 was substantially larger than the
56
7.0 percent decline that occurred during the Great
Trade Collapse (2008:Q2-2009:Q2). Taken together,
these facts indicate that oil imports played a relatively
modest role in the most recent U.S. trade contraction.
Having reviewed the role of trade in petroleum
goods, we now focus on manufactured goods, which
consist of durables and nondurables. In contrast to the
modest decline in oil imports, U.S. imports of nonoil
goods imports fell by 24.3 percent over the period
2008:Q2-2009:Q2. Figure 10 plots U.S. trade of non
durable and durable goods. The decline in trade of
durable goods—for example, automobiles, washing
machines, and industrial machinery—was more severe
than the decline in trade of nondurable goods—for
example, clothing and food. During the Great Trade
Collapse (2008:Q2-2009:Q2), imports in nondurable
goods declined by 10.98 percent, while imports in
durable goods declined by 28.6 percent; over the same
period, U.S. exports in nondurable goods declined by
6.8 percent, while exports of durable goods declined by
24.3 percent. These findings suggest that an economic
model that hopes to successfully quantify the contribu
tions of various factors to the Great Trade Collapse
should be characterized by a unique role for trade in
durable goods.
Supply-side explanations
At the start of the 2008 economic crisis, policy
makers observed that trade was falling dramatically
and began to question the cause. Anything that reduces
trade by raising the cost of selling a good in a foreign
market is considered a supply-side cause. Two problems
immediately raised concern. First, given that the world
was in the midst of a financial crisis, policymakers ques
tioned whether there was difficulty in obtaining trade
finance. Second, policymakers questioned whether there
was a rise in trade protection—that is, increases in import
taxes or other government-sponsored barriers to trade.
2Q/2011, Economic Perspectives
for goods that are bought and sold
domestically because there is a high
U.S. imports of goods and services, 1965-2010
er risk of nonpayment. If a seller
gives merchandise to a domestic
billions of chained 2005 U.S. dollars, seasonally adjusted
purchaser and the purchaser does
not pay, the seller can take the pur
chaser to court. However, when the
seller and purchaser are in different
countries and the purchaser does
not pay, it can be costly for the seller
to get what is owed from the pur
chaser. To mitigate this problem,
banks can get involved in the pay
ment process for international sales.
Most trade—80-85 percent—
occurs without any formal financing
and/or insurance arrangements with
banks.15 Still, banks are involved in
such international trading activity.
An
“open account” payment is made
Source: Authors’ calculations based on data from the U.S. Bureau of Economic
Analysis, National Income and Product Accounts of the United States, from
by the purchaser’s (importer’s) bank
Haver Analytics.
to the seller’s (exporter’s) bank after
the purchaser receives the goods.
However, banks are not extending
loans or offering insurance under
FIGURE 9
open account transactions. This is
U.S. imports of petroleum and nonpetroleum goods, 1965-2010
the least secure method of payment
for a seller; hence, this method is
billions of chained 2005 U.S. dollars, seasonally adjusted
most frequently used between parties
that have a well-established, long
standing relationship. Because there
is no guarantee, verification, or in
surance supplied by a third party,
payment for merchandise on an
open account is the cheapest way
to process a transaction.
The remaining 15-20 percent
of world trade is financed through
“letters of credit,” “documentary
collections,” and similar products
provided by banks or other third
parties.16 These instruments, which
Nonpetroleum
I Petroleum
come in many varieties, are payment
methods in which a payment is re
Source: Authors’ calculations based on data from the U.S. Bureau of Economic
Analysis, National Income and Product Accounts of the United States, from
leased from the buyer’s (importer’s)
Haver Analytics.
bank to the seller’s (exporter’s)
bank after certain documents have
been presented to the buyer’s bank
that verify delivery of the merchandise. Different types
Financial difficulties associated with trade
of these payment methods involve different levels of
Before discussing trade finance with respect to the
verification. The cost of using these products generally
Great Trade Collapse, it is useful to review the differ
increases as the level of verification becomes more
ent types of trade finance. The payment methods for
stringent, with letters of credit being more stringent and
internationally traded goods differ from those used
FIGURE 8
Federal Reserve Bank of Chicago
57
FIGURE 10
U.S. imports and exports of durables and nondurables
A. Imports
billions of chained 2005 U.S. dollars, seasonally adjusted
------- Nondurables
B. Exports
billions of chained 2005 U.S. dollars, seasonally adjusted
-------
Durables
Note: The shaded areas indicate official U.S. periods of recession as identified by the National Bureau of Economic Research.
Source: Authors’ calculations based on data from the U.S. Bureau of Economic Analysis, National Income and Product Accounts
of the United States, from Haver Analytics.
more costly than most other products. According to
the International Chamber of Commerce Banking
Commission (2010), the cost of commercial letters of
credit had increased during the recent financial crisis.
The price of a letter of credit varies according to factors
like the country of origin, the destination country, and
the industrial sector. The International Chamber of
Commerce survey evidence shows that prices of such
letters increased by as much as 300-400 basis points
over interbank lending rates during the height of the
financial crisis in the fall of 2008.17
To assess whether financing difficulties with these
payment instruments were a likely cause of the Great
Trade Collapse, we need information on the quantities
of financial instruments sold, their prices, and the vol
ume of trade. Two prominent surveys18 were undertaken
in early 2009 to try to fill in the gaps in policymakers’
knowledge about trade finance. While reports explaining
both surveys inform our understanding of the trade
finance situation during the crisis, both are thin on hard
statistics. The reports indicate that in the uncertain envi
ronment of the financial crisis, there was a relative in
crease in demand for more secure methods of payment,
such as letters of credit. However, because of the dif
ficulty in obtaining data on the number of open account
transactions involving merchandise trade,19 it is not
possible to evaluate how the proportions of unsecured
versus more secured methods of payment changed during
the crisis. While the surveys show that exporters sought
58
out more secure payment methods as the crisis worsened,
the total volume of letters of credit and documentary
collections fell. Presumably, this occurred because trade
fell. Lastly, according to the International Chamber of
Commerce Banking Commission (2010), refusals of
payments on minor technicalities increased through
out the crisis and remained high in early 2010. These
refusals were possible because even though letters of
credit promise payment when documents are presented,
a bank can refuse to make a payment if there are small
discrepancies in the paperwork that is filed.
To summarize, it appears that the cost of trading
goods internationally likely increased during the Great
Trade Collapse as exporters, worried about nonpayment,
began to use more secure and expensive methods of
payment. However, the precise magnitude of this cost
increase is not known.
What about other areas of trade financing? Letters
of credit and documentary collections are not the only
method of insuring payment by a foreign buyer. Export
credit insurance can be purchased by exporting firms
so that they are paid in the event of nonpayment by a
foreign purchaser. As the recent global recession
worsened, it appears that the use of export credit in
surance increased from roughly 9 percent of world trade
in 2008 to 11 percent of world trade in 2009.20 Claims
paid to insured customers by members of the Berne
Union, the leading international organization for export
credit and investment insurance, doubled from 2008
2Q/2011, Economic Perspectives
to 2009—from $1.1 billion to $2.4 billion. While this
is a substantial increase, it covers only a small percent
age of world merchandise trade (exports) in 2008, which
the World Trade Organization (2009) estimated at
$15.8 trillion.
A third way in which the financial system can af
fect international trade flows is through the provision
of trade credit. Recall that transactions on an open
account involve funds transfers between the buyer’s
bank and the seller’s bank, but do not involve a loan
from a bank. Rather, this type of sale is recorded as
a positive “accounts receivable” for the exporter and,
thus, is an informal loan from the exporter to the im
porter. The provision of trade credit is more common
in some industries than in others. For example, Chor
and Manova (2010) calculate the amount of trade credit
provided to buyers by suppliers in North America from
1996 through 2005.21 Industries such as transportation
equipment manufacturing, which has a North American
Industry Classification System (NAICS) code of 336,
and fabricated metal product manufacturing (NAICS
332) receive relatively more trade credit than industries
such as textile product mills (NAICS 314) or chemical
manufacturing (NAICS 325). During the financial crisis,
as the cost of borrowing money from a bank rose, it
would have become more expensive for exporting
firms to extend trade credit to their purchasers.
While the rising cost of trade credit affects all
firms that typically extend trade credit to their purchasers,
there are reasons to believe that the problem could have
been more severe for exporting firms. Domestic-marketoriented businesses as well as export-oriented businesses
often obtain working capital loans from banks to cover
the cost of purchasi ng inputs, paying workers, or renting
equipment. They repay these loans after receiving pay
ment from a buyer. For exported shipments, the time
lag between the shipment of goods and payment receipt
is 30-90 days longer than for domestic transactions.22
This means that working capital loans are especially
important for export-oriented firms.
To reiterate, an analysis of international trade pay
ment methods is not going to provide much important
information about whether the financial crisis had a
unique impact on trade. While survey data suggest that
costs of payment methods and export credit insurance
increased, precise quantitative data are not available
for economists to analyze. Economic research on the
role of finance in the Great Trade Collapse will likely
be more fruitful if it focuses on traditional credit instru
ments—such as working capital loans and trade credit.
Federal Reserve Bank of Chicago
The role of trade protection in the Great Trade Collapse
Before diving into a description of how trade
policy changed during the Great Trade Collapse, it is
useful to review the general trends in trade protection
leading up to September 2008. A dramatic reduction
in tariff rates and other nontariff barriers to trade
began with the end of World War II; this reduction,
combined with reductions in transportation and com
munication costs, led to dramatic increases in global
trade that outpaced the growth of global economic
activity for the past few decades. Under the auspices
of trade agreements like the World Trade Organization’s
GATT,23 most countries around the world have, to a
large degree, given up their unilateral authority to
raise trade barriers. Members of the WTO agree to
refrain from raising tariffs or imposing quotas above
certain “bound” limits in exchange for the same cour
tesy from other countries.
However, the GATT gives countries permission
to use some forms of trade protection under a variety
of special agreements or exceptional clauses. For ex
ample, a special tariff known as an antidumping duty
can be imposed on specific products imported from a
single country if a variety of economic criteria are met.
However, this type of country-specific trade restriction
has been found to be porous;24 if the United States
restricts imports of a product from Japan by using an
antidumping duty, another country like Germany will
simply increase its exports of that same product to the
United States, leading to, at most, a small reduction
in total U.S. imports of that product. Economists know
less about the effects of nontariff forms of trade pro
tection. Government intervention into markets, regula
tory changes, and changes in administrative procedures
or health or environmental policies can be subjected
to GATT disciplines if their trade-distorting effects
are large. These less transparent policies can be difficult
to identify, but organizations that run efforts like the
Global Trade Alert database25 have begun the difficult
task of compiling information about such policies and
then analyzing their effects.
Did we observe dramatic increases in trade barriers
at the time of the Great Trade Collapse? No. Growing
evidence suggests that, to date, trade protection has
been more muted than expected and its trade-distorting
effect has been mild at best.26
Figure 11 presents the recent state of U.S. trade
protection activity under the antidumping duty. This
is a special duty that the United States can use to restrict
imports when a domestic industry is suffering injury—
typically measured as reductions in employment and
capacity utilization27 as well as reduced profitability—
by reason of “dumped,” or unfairly priced, imports.
59
FIGURE 11
U.S. antidumping activity, 1979-2009
Antidumping investigations (left-hand scale)
| Antidumping import restrictions (left-hand scale)
------ U.S. imports (right-hand scale)
Source: Authors’ calculations based on data from Bown (2010).
We plot of the frequency of newly initiated antidump
ing investigations and new antidumping import restric
tions in the United States from 1979 through 2009, as
well as the level of U.S. imports in billions of chained
2005 U.S. dollars.
In figure 11, the height of the light blue bar mea
sures the number of new investigations that the U.S.
government conducted into allegations of unfairly priced
imports, while the height of the dark blue bar records
how many of these investigations ultimately resulted
in trade-reducing antidumping duties. The unit of ob
servation is an investigation held into or trade restriction
imposed against an individual country that exports to
the United States.28 The tallest bar on the graph is in
1992; the light blue bar indicates that the U.S. govern
ment conducted 94 investigations into allegations of
dumping, and the dark blue bar indicates that 39 of these
investigations found evidence of dumping and, con
sequently, resulted in trade-restricting import duties.
Superimposed over this graph of antidumping
activity, the black line indicates the real volume of
U.S. imports, in billions of chained 2005 U.S. dollars.
It shows a strong and steady increase in U.S. imports
that declined quite dramatically in 2008 and 2009.
We can clearly see that for the United States, there
were increases in both the number of antidumping
investigations and the number of investigations that
resulted in new antidumping duties in 2008 and 2009
60
relative to the pre-crisis years of 2005 and 2006.
Moreover, these did occur as U.S. imports were falling.
However, this rise in antidumping protection is consid
erably smaller than the jumps in trade protection during
earlier recessions. Previous spikes in antidumping activ
ity coincided with the period of the strong U.S. dollar
in the mid-1980s, the wake of the 1990-91 recession,
and, most recently, the 2001 recession. Further, when
we compare the number of antidumping investigations
that resulted in dudes to the total volume of U.S. imports,
we see that the fraction of U.S. imports subject to an
tidumping duties appears to be quite low in 2008-09.
Our principal observation here is that antidumping activi
ty, which has been the most popular method of trade
protection in the United States since 1980, did not
increase significantly during the crisis.
Figure 12 depicts the same information for Canada
over the period 1985-2009. The key observation is
that the pattern is quite similar to that in the United
States. There was a small uptick in activity in 2008
over 2007, but the use of antidumping trade restrictions
was quite modest by recent historical standards. Further,
when we compare the recent use of antidumping duties
to the total volume of Canadian trade, it appears to
be trivial.
How did trade protection evolve during the Great
Recession? For most countries, there have not been
substantial increases in explicit border measures like
2Q/2011, Economic Perspectives
FIGURE 12
Canadian antidumping activity, 1985-2009
Antidumping investigations (left-hand scale)
| Antidumping import restrictions (left-hand scale)
------ Canadian imports (right-hand scale)
Source: Authors’ calculations based on data from Bown (2010).
tariffs or quotas. If countries are changing their domestic
regulations, administrative procedures, or health and
safety standards in ways that discriminate against im
ported goods (and thus have trade-restricting effects),
these types of measures can be difficult for business
people and policymakers to observe. Further, even
when a potentially trade-distorting policy like the
“Buy American” provision of the 2009 U.S. stimulus
bill is well known, its trade impact can be difficult for
economists to measure.
While nontariff barriers could have a negative
effect on trade, existing evidence from initiatives like
the Global Trade Alert project suggests that the use of
these policies has been restrained. The effect of major
industrial policy initiatives on trade (for example, the
General Motors bailout in the United States) has yet
to be formally analyzed by researchers.
Summarizing the hypotheses behind
the Great Trade Collapse
To summarize, we posited three leading hypotheses
for what caused the collapse: 1) a decline in aggregate
demand for all goods, including imports; 2) difficulties
in obtaining trade finance; and 3) rising trade barriers.
A quick analysis has suggested that the fall in aggregate
demand can explain about half of the decline of imports
into the United States. Our review of the changing com
position of imports suggests that to fully understand
Federal Reserve Bank of Chicago
how declining demand affected trade during the Great
Recession, a richer economic analysis that examines the
structure of production and the composition of consump
tion and trade is needed. With regard to trade finance,
we explained why the lack of data on open account
transactions makes it difficult to draw conclusions from
the available data on payment methods for international
trade. More fruitful avenues of research would examine
how working capital loans and the provision of trade
credit could have been mechanisms through which
the recent global financial crisis reduced trade flows.
Finally, with regard to trade protection, it seems that
changes in traditional border barriers were not behind
the trade collapse. In fact, governments’ willingness
to refrain from trade restrictions allowed the trade re
covery to progress swiftly. However, as high unemploy
ment persists in much of the industrialized world, the
calls for more trade protection and accusations of cur
rency manipulation have been rising. Interestingly, in
figure 11, the United States’ aggressive use of antidump
ing duties associated with the 1990-91 recession peaked
not during the recession itself, but in 1992, as high
unemployment persisted with the United States’ “job
less recovery.”
Recent research on the Great Trade Collapse
A large literature is emerging on the causes of
the Great Trade Collapse. Here, we summarize and
61
review four important contributions. Each of these
papers uses a different methodology and emphasizes
different aspects of the trade collapse. From them, we
can glean a composite picture of the collapse and be
gin to quantify the contributions of underlying causes.
This, in him, will guide us in assessing the policy actions
undertaken by the G-20. Recall, as a starting point,
that the simple trade elasticity analysis from table 2
(p. 55) indicates that the decline in U.S. aggregate de
mand explains around 35-50 percent of the United
States’ import collapse. What have other researchers
learned about the causes of the Great Trade Collapse?
Levchenko, Lewis, and Tesar (2010)
Levchenko, Lewis, and Tesar (2010) ask how im
portant was declining aggregate demand in explaining
the collapse of trade in the United States. Their analy
sis uses highly disaggregated data on trade flows and
finds that the greatest declines occurred in sectors in
which vertical production linkages29 are most important.
In contrast, they find little to no evidence that trade
financial difficulties were behind the United States’
trade collapse.
Their paper proceeds in three distinct phases. First,
they present data documenting the scale and industrial
composition of the Great Trade Collapse in the United
States. Second, they conduct a “trade wedge” analysis
of macroeconomic data (which is discussed further in
the next paragraph). Third, finding that a large portion
of the United States’ trade collapse cannot be explained
by declining aggregate demand, they examine other
possible causes of the collapse. They undertake a crosssectional industry analysis of 1) vertical linkages among
firms, 2) financial constraints, and 3) differences in
the composition of trade and domestic demand to
identify the most important causes of the Great Trade
Collapse outside of falling aggregate demand.
The “trade wedge” analysis is similar to the predic
tions made using trade elasticities, which we presented
earlier. The idea is to determine the “wedge,” or differ
ence, between the actual decline in trade and the decline
in trade that is due to changes in demand and changes
in relative prices. The authors begin with a standard
import demand function that relates changes in imports
to changes in the price of domestic goods relative to
the price of imported goods and to changes in consump
tion and investment in the importing country. This
function assumes that domestically produced and for
eign goods are imperfect substitutes for one another
and that the amount of imports increases as the price
of domestically produced goods rises relative to the
price of foreign goods. Further, imports increase as
the total amount of domestic consumption and invest
ment increase. Import demand is given by:
62
l)
y
vap
,>■ )+(D),
where D = C +1; yf is the change over time in the
logged level of imports; s is the elasticity of substitution
between domestic and foreign goods; P is the change
in the log of domestic prices; pf is the change in the
log of import prices; and D is the change in the log
of total consumption and investment in the importing
country. Following previous research, the authors
assume that s is equal to 1.5. They use this equation
to predict the magnitude of the decline in U.S. imports
over the period 2008:Q2-2009:Q2, given the actual
quarterly data on changes in relative prices and changes
in U.S. consumption and investment from this period.
There are two important distinctions between
this analysis and our analysis using trade elasticities.
First, Levchenko, Lewis, and Tesar (2010) include a
measure for relative prices. Inclusion of these price
measures should increase the predictive power of their
model relative to a trade elasticity analysis that only
examines changes in demand. Second, they assume
that the import elasticity with respect to income is
one, roughly half the magnitude of the empirical esti
mates reported in table 2 (p. 55). From their analysis,
Levchenko, Lewis, and Tesar find that their standard
import demand equation explains 60 percent of the
decline in imports. The wedge is a 40 percent differ
ence between the actual decline in imports during this
period and the decline in imports predicted by their
import demand equation.
To demonstrate the uniqueness of the Great Trade
Collapse as an economic phenomenon, Levchenko,
Lewis, and Tesar (2010) calculate the size of the wedge
for every year-over-year change30 since 1968. They
find that the average wedge has been 2.9 percent since
then. More recently, this import demand equation has
improved in its ability to explain the behavior of imports.
Since 1984, the average wedge has been 1.6 percent.
What this means is that, while changes in relative prices
and in domestic demand can explain almost all of the
change in U.S. imports in a typical year, the wedge of
40 percent during the Great Trade Collapse was an
aberration that, at first blush, is hard to explain.
Faced with this puzzle, Levchenko, Lewis, and
Tesar (2010) refine their analysis of the trade wedge
to look as subsectors of the economy. They calculate
the trade wedge for nonoil imports, durable goods,
consumption goods, and investment/capital goods.
The trade wedges for consumer goods (which represent
around 20 percent of U.S. imports) and for investment/
capital goods (which also represent around 20 percent
of U.S. imports) are small, -6.4 percent and -10 percent,
2Q/2011, Economic Perspectives
respectively. For these sectors, the fall in demand and
change in the relative prices explain almost all of the
decline in imports. In contrast, the trade wedge for dura
ble goods is a sizable -21 percent. While substantial,
this is considerably smaller than the aggregate wedge
of 40 percent. Thus, controlling for the composition
of the trade flow can help explain some of the puzzle,
and the authors conclude that the unusual behavior of
trade in intermediate inputs and durable goods must
be behind some of the unexplained portion of the Great
Trade Collapse.
Using industry-level data on the percent change in
the flow of imports into the United States from June 2008
through June 2009, Levchenko, Lewis, and Tesar (2010)
explore three hypotheses for what caused the Great
Trade Collapse. First, they study the role of vertical
linkages in production. Did goods that are used inten
sively as intermediate inputs in production experience
large percentage drops in exports and imports? Second,
they ask how financial constraints affected trade. Spe
cifically, they analyze whether sectors that extend or
that intensively utilize trade credit experienced differ
ential changes in their trade flows relative to sectors
that do not. Finally, they investigate the role of trade’s
industrial composition. Was the United States’ trade
collapse unusually large because it was concentrated
in goods purchased or sold by sectors that were espe
cially hard hit during the Great Recession?
To test these hypotheses, Levchenko, Lewis, and
Tesar (2010) use data on approximately 450 sectors in
the United States to estimate the following equation:
2)
= a+ pCT/zIT?, + yW,. + e,..
In this equation,
is the percent change in a
trade flow from June 2008 to June 2009, CHAR is a
measure of the industrial sector that will test one of
the hypotheses (vertical linkages, trade credit, or sectorlevel industrial production), and X. is a vector of
industry-specific control variables.
To test the vertical linkages hypothesis, the authors
create a measure that captures the intensity with which
each good is used as an intermediate input in produc
tion. They use the input-output matrix from the U.S.
Bureau of Economic Analysis to calculate the average
amount of a commodity input, z, used to produce a U.S.
dollar’s worth of output in all downstream industries,
j. The authors find that goods used intensively as in
termediate inputs experienced larger percentage drops
in imports and exports.
Turning to the hypothesis that tight financial con
ditions contributed to the Great Trade Collapse, the
authors calculate two measures of trade credit intensity
Federal Reserve Bank of Chicago
in an industry. Using data from the Compustat North
America database, they calculate the amount of credit
extended to a firm by its suppliers as the median ratio
of accounts payable to cost of goods sold. A second
measure captures the amount of credit a firm extends
to its customers—specifically, this is measured as the
median ratio of accounts receivable to sales.31 For
example, if a firm that typically extended trade credit
to its buyers had difficulty obtaining working capital
from banks during the financial crisis, that firm might
cease to offer trade credit. Consequently, that might
have led to a decline in U.S. exports.
The authors find no evidence that trade flows fell
more in sectors that typically either extend or receive
trade credit. An examination of changes over time in
the ratio of accounts payable to cost of goods sold
and the ratio of accounts receivable to sales for firms
in the Compustat database over the periods 20002009:Ql and 2004:Ql-2009:Ql, respectively, reveals
that the contractions in trade credit during the financial
crisis were relatively small. This supports the authors’
conclusion that difficulties in obtaining trade credit
were not a major factor behind the Great Trade Collapse.
This analysis does not disprove the idea that tight finan
cial conditions could have contributed to the trade
collapse. Rather, the analysis indicates that, after con
trolling for other characteristics, sectors that regularly
require upfront payments for inputs and sectors that
regularly ship inputs to buyers in advance of payment
experienced similar declines in trade.
Finally, to test the hypothesis that the Great Trade
Collapse occurred because of compositional differences
between domestic output and trade, the authors exam
ine the relationship between the cross-sectional con
traction in output and the cross-sectional contraction
in trade. For this analysis, an industry-specific measure
of industrial production is used as the variable CHAR.
in equation 2. Compositional differences do account
for some of the Great Trade Collapse, according to
Levchenko, Lewis, and Tesar (2010). In an examination
of cross-sectional differences, imports and exports
contracted more in sectors in which U.S. industrial
production contracted more. Imports in durable goods
sectors contracted 9.2 percentage points more than
imports in nondurable goods sectors.
In summary, Levchenko, Lewis, and Tesar (2010)
first quantify that approximately 60 percent of the Great
Trade Collapse is due to the contraction in domestic
demand associated with the Great Recession and to
changes in the relative price of imports to domestic
goods. They then analyze cross-sectional changes in trade
flows and conclude vertical linkages and composi
tional differences between domestic production and
63
trade were important contributing factors to the Great
Trade Collapse. This partial equilibrium cross-sectional
approach does not lend itself to quantification of the
underlying causes of the collapse of aggregate U.S.
imports. However, from this empirical analysis, we
can see that a good economic model of the Great Trade
Collapse must include a distinction between nondurable
and durable goods and a careful modeling of inputs
and final goods.
Eaton et al. (2011)
Eaton et al. (2011) take a different approach to
studying the Great Trade Collapse. They complete an
empirical analysis on the Great Trade Collapse as a global
phenomenon. This paper begins with the observation
that the ratio of global trade to GDP declined by about
30 percent from 2008:Q2 through 2009:Q2. In con
trast to Levchenko, Lewis, and Tesar (2010), who are
agnostic about the underlying structure of the economy,
Eaton et al. (2011) build a structural model of the global
economy. They then use their model to reproduce the
Great Trade Collapse from possible causes. This
methodological approach has the additional benefit
of allowing the authors to quantify the contributions
of different factors to the Great Trade Collapse.
Eaton et al. (2011) begin with a standard gravity
model of trade among 23 countries. This workhorse
model of the international trade literature relates the
volume of trade between any two countries to the dis
tance between them.32 To the gravity model, they add
three production sectors—durable manufacturing,
nondurable manufacturing, and nonmanufacturing—
and a detailed input-output structure for each country.
The possible causes of the trade collapse are included
in the model as “shocks,” variables subject to exoge
nous changes in their value that can then propagate
throughout the model economy. In the Eaton et al.
(2011) model, there are four distinct types of shocks:
demand shocks, trade deficit shocks, productivity
shocks, and trade friction shocks.
In this paper, a demand shock, which is countryspecific, is a change in the share of final demand that
is spent on goods from each sector—durables, nondu
rables, or nonmanufacturing. In this setup, changes in
final investment activity or changes in durable inven
tories are captured by demand shocks. The equilibrium
in this model is a function of each country’s aggregate
trade deficit and its nonmanufacturing deficit. Because
the model is static, these trade deficits are treated as
exogenous shocks. Productivity shocks—which measure
how much of an output change cannot be explained
by changes in inputs of capital, labor, and materials—
and trade friction shocks—which capture all kinds of
changes in barriers to trade—are estimated from data
64
on sectoral producer price indexes and bilateral trade
shares at the sectoral level. The trade friction shocks
capture anything that changes individuals’ home bias33
in consumption, such as 1) changes in shipping costs,
2) changes in tariffs, 3) changes in nontariff barriers,
and 4) difficulties in obtaining trade finance. Further,
any reduction in imported inventories associated with
a large fixed cost of importing—as in Alessandria,
Kaboski, and Midrigan (2010) discussed later—would
also be captured by the trade friction shock.
The authors find that a decline in the demand for
durable manufactured goods explains 65 percent of
the decline in the global trade-to-GDP ratio during
this period. The decline in total demand for durable
and nondurable manufactured goods explains about
80 percent of the fall in the global trade-to-GDP ratio.
Finally, they find that increases in trade frictions
(difficulties with trade finance and rising trade protec
tion) reduced trade for China and Japan but had little
or no impact on other countries. How is it that Eaton
et al. (2011) find that 80 percent of the trade collapse
is due to the decline in demand, while a simple im
port demand analysis implies that declining demand
can explain only about half of the collapse? A key
difference between Eaton et al.’s (2011) analysis and
the import demand analysis in this article or that con
ducted by Levchenko, Lewis, and Tesar (2010) is that
Eaton et al. (2011) develop a richer model that incor
porates important features of the vertical structure of
trade and production. In their richer framework, a fi
nal demand shock in one country can fully propagate
itself through the demand for traded inputs into pro
duction of both durables and nondurables.
Chor and Manova (2010)
Both Levchenko, Lewis, and Tesar (2010) and
Eaton et al. (2011) examined demand and supply fac
tors as possible causes of the Great Trade Collapse,
and found that weak demand was quantitatively the
most important factor. A study by Chor and Manova
(2010) focuses on a supply-side cause by looking at
the availability of trade financing during the financial
crisis. When global credit markets froze, the market
for trade credit tightened, but not nearly as severely
as other markets. The paper concludes that tighter
trade financing conditions contributed to the collapse,
but this contribution was modest.
Chor and Manova (2010) ask how tight credit
affected trade volumes. Their empirical analysis of
the Great Trade Collapse focuses on whether countries
with higher borrowing costs exported less to the United
States during the crisis. Their paper exploits cross
country and intertemporal variation in the interbank
rate, the interest rate at which banks lend to one another,
2Q/2011, Economic Perspectives
to identify if tight financial conditions differentially
affected different countries’ monthly exports to the
United States. While the global nature of the financial
crisis meant that interest rates in different countries
followed a similar path throughout the crisis, Chor
and Manova use high-frequency data to capture small
differences in borrowing costs across countries and
over time. They hypothesize that countries with lower
interest rates should have experienced smaller declines
in the volume of their exports to the United States.
Consider Chor and Manova’s (2010) simplest
model—the relationship between U.S. imports from
different countries, designated i, in different three-digit
NAICS industrial sectors, designated k, at a monthly
frequency, t, as a function of the interbank lending
rate in country i over time.
3) inYikt-7lIBRATEit+y2DcnsisxIBRATEit + ^ + 8,,
where F,; is U.S. imports from country i in sector k in
month t, IBRATE.the interbank rate in country i and
month t, the variable D is a 0-1 indicator variable
equal to 1 in every month from September 2008 through
August 2009, the variable D,: is a full set of sectormonth fixed effects, and s.fe is an error term. The co
efficient y captures the effect of a change in the inter
bank rate on a country’s exports to the United States,
whereas the coefficient y2 captures the additional effect
of the interbank rate on a country’s exports to the United
States during the financial crisis. This formulation allows
for the possibility that the interbank rate might have
affected trade flows during the crisis in an unusual way.
From this simple model, Chor and Manova (2010)
find (in a specification that omits the crisis dummy)
that a 1 percent increase in the cost of bank financing
is associated with a 16 percent fall in that country’s
exports to the United States. However, after control
ling for industrial production in the exporting country,
the effect of the interbank rate on exports is generally
not significant. Thus, while there is some evidence
that tighter financial conditions, measured as a higher
economy-wide borrowing rate, was associated with a
lower level of exports to the United States, it is not
clear whether this decline in exports was caused by
tighter borrowing conditions or whether the tighter
borrowing conditions were simply correlated with
other adverse changes occurring in these exporting
economies during the crisis.
Chor and Manova (2010) then turn to a more re
fined question of whether sectors that are more reliant
on financing exported less to the United States during
the crisis. They exploit cross-sector dependence on
different types of external financing, together with
Federal Reserve Bank of Chicago
intertemporal changes in the interbank rate, to learn how
the financial crisis affected trade flows of different
types of goods. They estimate the following empirical
model on monthly imports into the United States:
4)7 In Y.,ikt = D.it + D,kt + D,ik + G,IBRATE
x FIN,k
r1
it
+ ff D
x
IBRATE
x
FIN, + e , .
Again, i indexes a foreign country, k indexes a threedigit NAICS sector, and t indexes time in months. The
key innovation in this expression, relative to equation 3,
is the inclusion of the variable, FINk, one of three
time-invariant measures of financial vulnerability. All
measures of financial vulnerability are constructed
from all publicly traded firms in the Compustat North
America database.34 The authors first calculate the
average value of the financial vulnerability variable
for each firm over the period 1996-2005. They then
use the median value of this average within a sector
as the sector’s time-invariant measure, FIN,.
The first measure of financial vulnerability that
Chor and Manova (2010) analyze is the external finan
cial dependence of a sector. External finance depen
dence is the fraction of total capital expenditures not
financed by internal cash flows from operations. Thus,
we might expect that sectors with high levels of this
variable would experience greater declines in trade
flows. The next measure they explore is asset tangi
bility—that is, the share of net plant, property, and
equipment in total book value. Because a firm with lots
of tangible assets can easily provide collateral for a loan,
one might expect that it is easier for these firms to ob
tain loans on advantageous terms. Finally, in a setup
similar to that of Levchenko, Lewis, and Tesar (2010),
Chor and Manova (2010) examine how access to buyersupplied trade credit affects cross-country exports at
the sectoral level. In their analysis, the change in ac
counts payable relative to the change in total assets
measures a sector’s access to buyer-supplied trade credit.
Chor and Manova (2010) find evidence that sup
ports the idea that financial difficulties contributed to
the Great Trade Collapse in the United States, but the
empirical support for this conclusion is not robust
across all specifications of their models. Overall, they
find that 1) sectors that are more reliant on external
finance had a slightly weaker export performance,
2) sectors with relatively more tangible assets exported
relatively more, and 3) sectors that routinely receive
trade credit from buyers experienced smaller declines
in their exports to the United States.
More specifically, when the fraction of total capital
not financed by internal cash is used as the measure
65
of financial vulnerability in equation 4, the coefficient
[> is identified from the variation in financial depen
dence across industries within a given country-month,
the variation in the cost of credit across exporting
countries in a given sector-month, and the variation
in the cost of credit over time within a given country’s
sector. The coefficient [V, relies on the same sources
of variation in the data for the months of the world
wide financial crisis. Empirically, the authors found
that P2 was negative and precisely estimated in almost
all specifications, but that estimates of (i were not
statistically different from zero. This suggests that dur
ing the financial crisis, high interest rates tended to
depress U.S. imports in financially vulnerable sectors.
With regard to the specifications that used the level
of tangible assets as the financial variable, recall that
a sector with more tangible assets should be less sen
sitive to worsening credit conditions because any loan
it requests can be collateralized by its tangible assets.
Thus, the authors hypothesize that both p and P2 should
be positive. In fact, they find that |> is positive in all
specifications, but statistically significant in only the
regression that omits the crisis dummy interaction
term. Further, P2 is positive in almost all specifications,
indicating that this effect was stronger during the finan
cial crisis. Thus, exporting firms that faced high borrow
ing costs performed better if they were in sectors with
relatively high levels of tangible assets.
Lastly, Chor and Manova (2010) consider the
role of trade credit in explaining the Great Trade Collapse.
These results are most directly comparable to those of
Levchenko, Lewis, and Tesar (2010), but the two papers
use different measures of trade credit.35 As stated pre
viously, one measure of financial vulnerability used
by Chor and Manova (2010), buyer-supplied trade credit,
is the change in accounts payable divided by the change
in total assets. This ratio measures how much credit
American purchasers in these sectors extend to foreign
exporters. The positive coefficient estimate on p indi
cates that countries with high interbank rates exported
relatively more in sectors in which American buyers
typically extend high levels of trade credit. The positive
coefficient estimate on P2 indicates that this effect be
came more pronounced during the crisis. This suggests
that financial constraints did exacerbate the collapse
of trade.
But how do we reconcile the different findings on
trade credit in Chor and Manova (2010) and Levchenko,
Lewis, and Tesar (2010)? The two papers exploit dif
ferent sources of variation in trade flows. Levchenko,
Lewis, and Tesar (2010) look at differences in the
provision of trade credit across sectors within the
United States. Their analysis looks for differences in
66
import growth across sectors that are systematically
linked to differences in trade credit, but does not find
significant changes in imports that coincide with the
trade credit measure. In contrast, Chor and Manova
(2010) exploit cross-country variation in the cost of
financing within a sector. They compare sectors A and B
in countries 1 and 2, all of which export to the United
States. Their analysis finds that if sector A receives a
relatively high level of trade credit and the interbank
rate is relatively higher in country 1 than in country 2,
then the relative exports of sector A to sector B in
country 1 will be larger than the relative exports of
sector A to sector B in country 2. This more refined
analysis is able to capture the subtle effects of financial
difficulties that varied across countries and over time.
Finally, Chor and Manova (2010) conduct counterfactual simulations with their model to try to quan
tify how severe the Great Trade Collapse would have
been if central banks and national governments had
not intervened to lower borrowing costs around the
world. They estimate that U.S. imports in the most
financially vulnerable sectors would have been substan
tially lower after September 2008 without the aggres
sive reduction in interbank lending rates that occurred.
Alessandria, Kaboski, and Midrigan (2010)
A final important contribution exploring the causes
for the Great Trade Collapse is Alessandria, Kaboski,
and Midrigan (2010). They develop a quantitative
dynamic model of trade and production to analyze
the Great Trade Collapse in the United States. Their
approach is unique in that it focuses on a new channel
of trade dynamics—namely, the behavior of inventory
investment over the business cycle.
Consider the following stylized example that
Alessandria, Kaboski, and Midrigan (2010) present.
Suppose a firm would ideally hold three units of a
good in inventory for each unit that it sells. In other
words, the firm’s ideal inventory to sales ratio is three.
If a recession causes the firm’s sales to fall, its inventory-to-sales ratio will increase above its ideal level.
This would lead the firm to purchase fewer goods from
its supplier to hold in inventory in the next period. If
the supplier is a foreign firm and the domestic firm’s
inventories are all imported goods, then a decline in
the domestic firm’s final sales in one period will lead
to a more than proportionate reduction in its purchase
of imported inventory in the following period.
Alessandria, Kaboski, and Midrigan (2010) formally
assess the role of inventory investment during the Great
Trade Collapse by integrating a partial equilibrium
model of trade and inventory adjustment into a twocountry general equilibrium model of trade. The key
2Q/2011, Economic Perspectives
feature of this model is that if transaction frictions are
higher for imported inventories than domestic inven
tories (that is, those purchased from domestic partners)
so that domestic producers with imported inventories
target a higher inventories-to-sales ratio, then any shock
that causes final sales to fall will have a larger effect
on imported inventories than on domestic inventories.
Alessandria, Kaboski, and Midrigan calibrate their model
to U.S. data and find that their model with inventory
decumulation generates dynamic patterns for production,
trade, and inventories that are quantitatively similar
to those observed during the Great Trade Collapse.
A particularly good feature of this model is that the
dramatic collapse in imports is followed by a sharp
recovery, similar to what we have observed for the
recovery following the Great Trade Collapse.
Conclusion
The collapse in international trade between the
second quarter of 2008 and the second quarter of 2009
is one of the most dramatic features of the Great
Recession. This collapse in world trade of over 17 per
cent from peak to trough was massive, not only in terms
of its U.S. dollar value but also by historical standards.
The G-20 leaders responded to this dramatic decline
in trade with three distinct policy initiatives—1) fiscal
stimulus to support aggregate demand, 2) trade finance
initiatives, and 3) promises to refrain from new trade
barriers. To assess the likely impact of these policies,
we explored in this article three main possible causes
of the Great Trade Collapse—namely, 1) declining
demand, 2) financing difficulties, and 3) rising trade
barriers. Economists have proposed several hypotheses
to explain the Great Trade Collapse; in addition to the
three already mentioned, some have posited the fol
lowing as possible contributing factors: differences in
the composition of trade and domestic output and the
behavior of imported inventories.
Federal Reserve Bank of Chicago
Research suggests that declining demand can
explain between 35 percent and 80 percent of the de
cline in trade over the period 2008:Q2-2009:Q2. The
analysis we perform in this article estimates that de
clining aggregate demand explains 35-50 percent
of the Great Trade Collapse. With regard to the recov
ery, our analysis finds a quantitatively larger puzzle;
rising aggregate demand explains only 25-40 percent
of the recovery in imports. The decline in aggregate
income is able to explain a larger fraction of the decline
in trade in a more sophisticated model that accounts
for differences between durable, nondurable, and non
manufacturing output, as well as the vertical structure
of production. The conclusion that declining demand
was the major cause suggests that of all the policy ac
tions, fiscal stimulus likely had the largest impact on
the trade recovery.
There is some evidence that financing difficulties
contributed to the Great Trade Collapse, but the pre
cise quantitative significance of financial factors is
difficult to assess. The G-20’s announcement in the
second quarter of 2009 that it would ensure the avail
ability of $250 billion for trade finance coincided with
the nadir of the Great Trade Collapse. However, we
cannot conclude from the coincidence in timing that
government aid with trade finance caused the trade
recovery. It likely had a positive impact that was dwarfed
by the positive impact of the economic recovery.
There is almost no evidence that trade policy
barriers rose during the period of trade collapse and
recovery. Historical experience with trade protection
ism teaches us that the trade collapse would almost
certainly have been worse if policymakers had responded
to the crisis by erecting new barriers to trade. Further,
it seems that the dramatic demand-driven trade recovery
was only possible because there were no trade barriers
in place to impede it.
67
NOTES
’Our calculation is based on data from the Organisation for Economic
Co-operation and Development’s Main Economic Indicators, in
which world trade in goods and services is defined as the sum of
world exports in goods and services and world imports in goods
and services divided by two; data are from Haver Analytics.
2For a complete list of G-20 nations, see www.g20.org/about_
what_is_g20.aspx.
from SWIFT (Society for Worldwide Interbank Financial
Telecommunication)—a private provider of electronic financial
messaging services. However, it is not possible to identify open ac
count transactions for merchandise trade by using SWIFT data be
cause there is no SWIFT message code uniquely specified for
payment for a sale of goods or services. Open account transactions
for goods are classified under the same SWIFT code as foreign ex
change sales. For more on SWIFT, see www.swift.com/about_
swift/press_room/SWIFT_for_media_July_2010.pdf.
3Group of Twenty (2008).
4Group of Twenty (2009), paragraph 22.
5Ibid., paragraph 6.
6See, for example, Alessandria, Kaboski, and Midrigan (2010);
Chor and Manova (2010); Eaton et al. (2011); and Levchenko,
Lewis, and Tesar (2010).
7For more on the GATT/WTO system, see www.wto.org/english/
thewtoe/whatis_e/tif_e/fact4_e .htm.
^Deardorffs ’ Glossary ofInternational Economics refers to this
phenomenon as “fragmentation.” Both vertical specialization and
fragmentation refer to “the splitting of production processes into
separate parts that can be done in different locations, including in
different countries” (Deardorff, 2010).
9We include all countries that have real quarterly trade and GDP
data series available—27 OECD countries, Brazil, and India meet
our criteria (see figure 6 for a complete listing).
10According to the National Bureau of Economic Analysis, the
Great Recession occurred in the United States in
2007:Q4-2009:Q2.
^International Chamber of Commerce Banking Commission
(2010), p. 46.
21Chor and Manova (2010) use data on all publicly traded firms
in the Compustat North America database to calculate the average
measure of the ratio of the change in accounts payable to the change
in total assets for each firm from 1996 through 2005. They then
take the median value across all firms in a three-digit North American
Industry Classification System (NAICS) industry as the industry’s
measure of trade credit.
22See Chor and Manova (2010), p. 3. Djankov, Freund, and
Pham (2010) present survey data from 98 countries, indicating
that the average time for a standardized container of merchandise
to be transported from a factory floor and cleared for export from
a country is 30 days.
23See note 7 for more on the GATT and WTO.
24Prusa (2001).
25For more on this database, which is coordinated by the Centre
for Economic Policy Research based in London, see
www.globaltradealert.org.
26See Evenett (2010).
"See Miroudot, Lanz, and Ragoussis (2009), table 7, p. 48.
12A vertically integrated international economy is one in which
supply chains cross international borders.
13Economists also estimate the responsiveness of trade to changes
in the prices of imported goods and services relative to domestically
produced ones. These estimates are referred to as trade elasticities
with respect to prices.
14See Crane, Crowley, and Quayyum (2007) for a detailed discussion
of trade elasticities.
"International Chamber of Commerce Banking Commission
(2010), p. 18.
"The U.S. Department of Commerce, International Trade
Administration (2008) provides a clear introduction to the
payment methods used in international trade.
"International Chamber of Commerce Banking Commission
(2010), p. 42.
"The findings from these two surveys are reported and analyzed
in the International Chamber of Commerce Banking Commission
(2009) and the International Monetary Fund and Bankers’
Association for Finance and Trade (2009).
"Data on the number of transactions that took place using
letters of credit or documentary collections can be obtained
68
27 “Capacity utilization is a ratio of a manufacturer’s actual produc
tion to their full production capability during [a specific time period],”
states the U.S. Census Bureau; see www.census.gov/manufacturing/
capacity/definitions/index. html.
28The amount of trade covered by each unit of observation varies
considerably—with some investigations covering only a portion
of a specific tariff line and others covering hundreds of tariff lines.
That said, this legally defined unit of observation has been used
consistently since 1980 and is a useful measure for looking at long
term trends.
"Measures of an industry’s downstream vertical linkages capture
the intensity with which the output of an industry is used as an
intermediate input by other sectors. Measures of an industry’s
upstream vertical linkages capture the intensity with which that
industry uses intermediate inputs. (See Levchenko, Lewis, and
Tesar, 2010, p. 14.) Deardorffs ’ Glossary ofInternational Economics
defines an intermediate input as “an input to production that has
itself been produced and that, unlike capital, is used up in produc
tion” (Deardorff, 2010).
30For example, they calculate the change between the first quarter
of year t and the first quarter of year t - 1, the second quarter of
year t and the second quarter of year t - 1, and so on.
3’Both measures in Levchenko, Lewis, and Tesar (2010) first take
the median value of the variable for each firm in the sample between
2000 and 2008. Next, they take the median of the value across all
firms to use as the industry’s measure of trade credit intensity.
2Q/2011, Economic Perspectives
32The term gravity model comes from the observation that trade
volumes tend to increase as the distance between any two countries
decreases, similar to the force of gravity between two objects in
creasing as the distance between the objects decreases. In addition,
this term is also based on the observation that the trade volume for
an economy grows larger as the size of an economy increases,
analogous to the gravitational pull of an object becoming larger as
its mass increases.
34Note that Chor and Manova (2010) are assuming that the financial
vulnerability in foreign industrial sectors is identical to that of the
same sectors in North America.
35The measures used in Levchenko, Lewis, and Tesar (2010) are
described in detail on p. 63 and in note 31.
33From Deardorffs ’ Glossary ofInternational Economics'. Home
bias is “a preference, by consumers or other demanders, for prod
ucts produced in their own country compared to otherwise identical
imports” (Deardorff, 2010).
REFERENCES
Alessandria, George, Joseph P. Kaboski, and
Virgiliu Midrigan, 2010, “The Great Trade Collapse
of 2008-09: An inventory adjustment?,” IMF Economic
Review, Vol. 58, No. 2, December, pp. 254-294.
Bown, Chad P., 2010, Global Antidumping Database,
World Bank Group, available at http://econ.worldbank.
org/ttbd/gad/.
Cardarelli, Roberto, and Alessandro Rebucci, 2007,
“Exchange rates and the adjustment of external im
balances,” in World Economic Outlook: Spillovers
and Cycles in the International Economy, International
Monetary Fund, report, April, pp. 81-120.
Chinn, Menzie D., 2004, “Incomes, exchange rates
and the U.S. trade deficit, once again,” International
Finance, Vol. 7, No. 3, December, pp. 451-469.
Chor, David, and Kalina Manova, 2010, “Off the
cliff and back? Credit conditions and international
trade during the global financial crisis,” National
Bureau of Economic Research, working paper,
No. 16174, July.
Crane, Leland, Meredith A. Crowley, and Saad
Quayyum, 2007, “Understanding the evolution of
trade deficits: Trade elasticities of industrialized
countries,” Economic Perspectives, Federal Reserve
Bank of Chicago, Vol. 31, Fourth Quarter, pp. 2-17.
Crowley, Meredith A., 2003, “An introduction to the
WTO and GATT,” Economic Perspectives, Federal
Reserve Bank of Chicago, Vol. 27, Fourth Quarter,
pp. 42-57.
Deardorff, Alan V., 2010, Deardorffs’ Glossary’ of
International Economics, University of Michigan,
available at http://www-personal.umich.edu/~alandear/
glossary/intro.html.
Federal Reserve Bank of Chicago
Djankov, Simeon, Caroline Freund, and Cong S.
Pham, 2010, “Trading on time,” Review ofEconomics
and Statistics, Vol. 92, No. 1, February, pp. 166-173.
Eaton, Jonathan, Samuel Kortum, Brent Neiman,
and John Romalis, 2011, “Trade and the global
recession,” National Bureau of Economic Research,
working paper, No. 16666, January.
Evenett, Simon. J. (ed.), 2010, Tensions Contained ...
For Now: The 8th GTA Report, Centre for Economic
Policy Research, November, available at www.
globaltradealert.org/sites/default/files/GTA8_0.pdf.
Freund, Caroline, and Diana Weinhold, 2000,
“On the effect of the Internet on international trade,”
International Finance Discussion Papers, Board of
Governors of the Federal Reserve System, working
paper, No. 693, December.
Group of Twenty, 2009, “The global plan for
recovery and reform,” leaders statement, London,
April 2, available at www.g20.org/Documents/
final-communique.pdf.
__________ , 2008, “Summit on financial markets
and the world economy,” declaration, Washington,
DC, November 15, available at www.g20.org/
Documents/g20_summit_declaration.pdf.
Hooper, Peter, Karen Johnson, and Jaime
Marquez, 2000, “Trade elasticities for the G-7 coun
tries,” Princeton Studies in International Economics,
Princeton University, working paper, No. 87, August.
Houthakker, H. S., and Stephen P. Magee, 1969,
“Income and price elasticities in world trade,” Review
ofEconomics and Statistics, Vol. 51, No. 2, May,
pp. 111-125.
69
Hummels, David, 2007, “Transportation costs and
international trade in the second era of globalization,”
Journal ofEconomic Perspectives, Vol. 21, No. 3,
Summer, pp. 131-154.
__________ , 2001, “Toward a geography of trade
costs,” Purdue University, mimeo, September.
International Chamber of Commerce Banking
Commission, 2010, “Rethinking trade finance 2010,”
report, Paris, April, available at www.iccwbo.org//
uploadedFiles/Rethinking_Trade_Finance_2010.pdf.
__________ , 2009, “Rethinking trade finance 2009:
An ICC global survey,” report, Paris, No. 470-1120
TS/WJ, March 31, available at www.iccwbo.org/
uploadedFiles/ICC_Trade_Finance_Report.pdf.
International Monetary Fund and Bankers’
Association for Finance and Trade, 2009, “IMFBAFT Trade Finance Survey: A survey among banks
assessing the current trade finance environment,”
report (based on survey conducted by FImetrix),
Washington, DC, March.
Levchenko, Andrei A., Logan T. Lewis, and Linda
L. Tesar, 2010, “The collapse of international trade
during the 2008-2009 crisis: In search of the smoking
gun,” IMF Economic Review, Vol. 58, No. 2, December,
pp. 214-253.
Miroudot, Sebastien, Rainer Lanz, and Alexandras
Ragoussis, 2009, “Trade in intermediate goods and
services,” OECD Trade Policy Working Papers,
Organisation for Economic Co-operation and
Development, No. 93, November.
Prusa, Thomas. J., 2001, “On the spread and impact
of anti-dumping,” Canadian Journal ofEconomics,
Vol. 34, No. 3, August, pp. 591-611.
Subramanian, Arvind, and Shang-Jin Wei, 2007,
“The WTO promotes trade, strongly but unevenly,”
Journal ofInternational Economics, Vol. 72, No. 1,
May, pp. 151-175.
U.S. Department of Commerce, International
Trade Administration, 2008, Trade Finance Guide:
A Quick Referencefor U.S. Exporters, Washington, DC:
Manufacturing and Services Office of Finance, April.
World Trade Organization, 2009, World Trade
Report 2009: Trade Policy Commitments and
Contingency Measures, report, Geneva, Switzerland,
available at www.wto.org/english/res_e/booksp_e/
anrep_e/worldtradereportO 9_e. pdf.
Yi, Kei-Mu, 2003, “Can vertical specialization
explain the growth of world trade?,” Journal of
Political Economy, Vol. Ill, No. 1, February,
pp. 52-102.
Levinson, Marc, 2006, 77?e Box: How the Shipping Con
tainer Made the World Smaller and the World Economy
Bigger, Princeton, NJ: Princeton University Press.
70
2Q/2011, Economic Perspectives
How do private firms use credit lines?
Sumit Agarwal, Souphala Chomsisengphet, and John C. Driscoll
Introduction and summary
Companies borrow from investors for a variety of reasons.
For example, current sales revenue may not be enough
to pay suppliers or employees; companies may wish to
make long-term investments by buying new equipment
or constructing new buildings; or they may want to have
access to credit to deal with unforeseen circumstances.
Large companies have a wide menu of choices for bor
rowing funds, including issuing new stock or bonds.
Small companies tend to have a smaller set of options.
Because such companies may also be younger than large
companies and, thus, have a shorter track record, or be
cause they may be more reliant on the performance of
a small number of key employees, these firms will face
more difficulty in conveying their value to the broad class
of investors who participate in the bond or stock markets.
Small firms are thus often privately held (that is,
their stocks are not traded on public exchanges). These
privatefirms likely rely on bank loans for much of their
borrowing, as banks may be better able to spend the
resources to investigate the firms’ prospects.1 Such small,
bank-dependent firms are vulnerable to problems in
the banking system. Indeed, a number of researchers
argue that monetary policy and other economic shocks
that impact the supply of credit flow through banks to
bank-dependent firms.2
Although banks make many traditional spot loans,
in which the whole amount of the loan is provided to
the firm, much business lending takes the form of a credit
line, also known as a Zon« under commitment. In a loan
under commitment, the bank agrees to provide funds to
the firm as needed up to a pre-specified limit, at mutu
ally agreed-upon terms and over a fixed period. As of
the end of the second quarter of 2010, commercial banks
held $1.1 trillion in commercial and industrial loans on
their books, but had about $2 trillion in unused com
mitments (that is, the portion of the credit line not yet
used) on business credit lines.3
Federal Reserve Bank of Chicago
The market for loans under commitment is impor
tant because it represents a large portion of business
lending and the majority of small business finance. In
addition, loans under commitment may be one channel
through which monetary policy and credit shocks are
transmitted to the broader economy. However, a lack
of available data has made it difficult to study loans un
der commitment or small business lending more broadly
defined. Standard government data sources on banking,
such as the Reports of Condition and Income, also
known as the Call Reports (produced by the Federal
Financial Institutions Examination Council, or FFIEC),
or the Federal Reserve System’s weekly bank credit
data (H.8 statistical release), do not break out business
lending by the size of the borrowing firm. Publicly traded
firms are required to issue quarterly reports on their
balance sheet, including details of their financing, but
private firms do not have such requirements.
In this article, we use a panel data set from a
large bank to examine the behavior of loans under
commitment made to privately held firms. The data set
contains all of the characteristics of the credit lines and
all of the financial information about the firms that is
available to the bank.
As with any lending market, the interest rate on the
loan under commitment, the collateral and other re
quirements for the credit line, and the amount of the
Sumit Agarwal is a senior economist in the Economic Research
Department at the Federal Reserve Bank of Chicago.
Souphala Chomsisengphet is a senior economist at the
Office of the Comptroller of the Currency. John C. Driscoll
is a senior economist in the Divison ofMonetary Affairs at
the Board of Governors of the Federal Reserve System. The
authors would like to thank Jim Papadonis for his support
of this research project. They would also like to thank Mike
Fadil, Kristen Monaco, Nick Souleles, and participants at the
Midwest Economic Association Meetingsfor helpful comments.
They are grateful to Diana Andrade, Ron Massinger, and
Kathy Parugini for excellent research assistance.
71
line are jointly determined by the intersection of the
bank’s supply and the firm’s demand. In the absence
of further identifying assumptions, we will not know
whether these prices and quantities change over time
or differ across firms because of changes in factors driv
ing supply or factors affecting demand for these loans.
However, our data set provides us with information
on both the amounts of credit that firms requested and
the amounts granted. Restricting our analysis to those
cases in which the amount requested is equal to the
amount granted helps us to ensure that observed differ
ences in prices and quantities across firms reflect
differences in firms’ demand for credit rather than
differences in banks’ willingness to supply credit. Still,
no attempt to solve the problem of separating supply
and demand is perfect, and some of our results on the
determinants of credit demand may partly capture
factors that affect credit supply instead.
Economists have hypothesized a number of reasons
why companies might choose to borrow via credit lines
rather than spot loans, including the need to hedge against
the possibility of a sudden deterioration in their own
creditworthiness and a desire for flexibility to be able
to quickly take up new investment opportunities. We
look at some of the factors that affect these and other
reasons underlying the demand for lines under commit
ment. We find that increases in fees paid on the com
mitment and the interest rate charged to the firm lead
to large reductions in the size of lines obtained—in
other words, the demand curve does indeed slope down
ward with the cost of the loan. Increases in fees for
overcharging the lines raise line demand (as firms pre
sumably try to avoid such overcharges by borrowing
more at the outset). Increases in mean profit growth—
a proxy for future investment opportunities—lead to
very large increases in credit lines, while increases in
the volatility of profit growth or in cash flow (a source
of internal funds) cause, respectively, large and mod
erate decreases in the size of lines; these results sug
gest that access to funds for flexibility is an important
motive, as described in the model developed by Martin
and Santomero (1997). We find weak evidence against
models in which loans under commitment help firms
to hedge against the possibility that their own credit
ratings may decline; we estimate that the quantity of
credit demanded is negatively related to measures of
firm risk.
If firms do use credit lines to enhance their flexibility,
many of the same factors that affect their demand for
the size of the line will also affect their usage of the line.
Firms will not want to use all of their lines, as that
would leave them at risk of not being able to fund new
opportunities. We test this idea by examining whether
72
line utilization responds to the same variables that in
fluence line demand. With the exception of upfront fees,
all variables affect line utilization in the same way as
they do line size.
In the next section, we summarize the academic
literature on business credit lines. We then discuss
our data set and the setup for our estimation. Finally,
we present our results and discuss their implications.
The economics of loan commitments
When a firm takes out a loan under commitment (or
credit line), the bank commits to providing up to some
amount of credit to the firm over a specified period.
The firm is not obligated to take out the full amount of
the credit line at once and, indeed, usually does not
do so even over the entire duration of the contract.
The bank charges the firm for setting up the line (known
as the commitment fee); it may also charge other fees
or penalties if the firm exceeds the line limit or other
wise breaks the contract. Both spot loans and credit
lines usually require the firm to post collateral.
Firms face some trade-offs in choosing between
spot loans and credit lines. For example, the existence
of the commitment fee, holding everything else equal,
makes a credit line more costly to a firm than a spot
loan. The economics and finance literature provides
several competing views on the relative merits of spot
loans and loans under commitment and how firms
choose between them.
According to one view, loan commitments allow
firms to hedge against any deterioration in their own
creditworthiness over the period of the loan.4 If a firm
suffers such a deterioration, it may have trouble getting
a new spot loan. Having a partly unused line of credit
would provide the firm with needed funds in this case.
This option would only be available if the bank was
not able to use the deterioration as an excuse to cut
the size of the firm’s credit line.
A second body of work argues that loan commit
ments help private firms hedge against decreases in
the aggregate supply of credit, or credit crunches.5
Firms may be concerned that a decrease in the supply
of credit by the banking industry—such as what occurred
in the aftermath of the savings and loan crisis of the
early 1990s—will leave them less able to borrow. Of
course, a banking industry crisis may coincide with a
period of declining creditworthiness. Both of these
phenomena may have been at work during the recent
financial crisis. In the third quarter of 2008, commer
cial bank lending to businesses expanded rapidly while
the fraction of loan commitments unused dropped, sug
gesting that businesses were drawing down their credit
lines during a time when activity in other corporate
2Q/2011, Economic Perspectives
credit markets, such as that for commercial paper, was
rapidly diminishing.6
A third view contends that loan commitments are
attractive to both firms and banks because they help solve
information problems that make it difficult for firms to
borrow on the spot markets for loans or commercial
paper.7 According to this view, some firms may be par
ticularly difficult to value, perhaps because they have
assets that have illiquid markets or because the firms
are small and rely heavily on the work of a few key in
dividuals. Such firms will have difficulty borrowing
in the bond and commercial paper markets since it will
be difficult to convey the riskiness of the securities to
the broad class of investors who participate in such mar
kets. Banks are better able to investigate the quality of
the firm and monitor its behavior. Credit lines also
provide more protection to the bank than spot loans
because the bank may have the option of cutting the
unused portion of the line if circumstances change.
A final view argues that the relative speed and flexi
bility offered by credit lines enables firms to take advan
tage of investment opportunities they might miss if they
had to take the time to obtain approval for spot-market
loans (see Martin and Santomero, 1997). This flexibility
makes the extra costs (in the form of fees and higher
interest rates) of loans under commitment worthwhile
to the firm.
These reasons are not mutually exclusive; it is likely
that all of them contribute, to some degree, to devel
opments in the market for credit lines. The empirical
evidence on these explanations is a bit mixed, in part
because of the data availability difficulties alluded to
in the introduction. Also, with a variety of explanations,
it is difficult to estimate the contribution of any indi
vidual one (and many studies have focused on evalu
ating one of many possible explanations). Several authors
have found that macroeconomic developments in the
market for bank loans appear to affect the quantity and
price of loans, providing support for the second view:
Borrowers take out credit lines because they are con
cerned about decreases in the aggregate supply of credit.8
Shockley and Thakor (1997) find some evidence for
the third view: Borrowers that appear to be harder
to value (because they are less well known or have
assets that are difficult to value) tend to use credit lines
rather than other nonbank forms of finance, such as
commercial paper.
Ham and Melnik (1987) look at the determinants
of usage of credit lines (that is, conditional on having
obtained a loan under commitment, what fraction of
that loan is used). Using a sample of 90 nonfinancial
corporations, the authors find that credit line usage is
positively related to total sales, borrowed reserves, and
Federal Reserve Bank of Chicago
whether collateral is used to secure the loan; and it is
negatively related to interest rate costs (specifically, risk
premiums and commitment fees).
Much of this empirical work has attempted to
identify what determines banks’ willingness to supply
credit. The papers that have focused on the demand for
credit have used data on larger, publicly traded corpora
tions. As we discuss in the next section, our data allow
us to study smaller firms that are not publicly traded and,
we argue, to analyze demand for credit by these firms.
Data and empirical strategy
Data
Our unique data set comes from a large commercial
bank that issued lines of credit to both publicly traded
and private firms. For this article, we restrict our sam
ple to private firms with fewer than 500 employees.
Our data set has independently audited quarterly balance
sheet data on the firms from the second quarter of 1998
through the fourth quarter of 2002 and monthly loan
performance information from the first quarter of 2001
through the fourth quarter of 2002.
Tables 1 and 2 provide some summary statistics for
the firms in our sample. The top panel of table 1 gives
the distribution of firms across industries and the bot
tom panel gives the distribution across geographical
locations. The firms are distributed across seven broadly
defined classes of industry, ranging from manufactur
ing to retail and wholesale trade to services, and are
located in five northeastern states.
Table 2 provides means and standard deviations
(a measure of dispersion) on other firm characteristics
and balance sheet information. The mean age of the
firms is about ten years. The firms on average hold
just above $2 million in total assets and have about
$630,000 in working capital. The firms in our sample
have relatively robust annual growth rates of profits and
sales, of about 22 percent and 25 percent, respectively.
On a scale of 1 to 8, with 1 being the least risky, the
average firm receives a rating of about 5. The remain
ing entries in the table are characteristics of the firms’
credit lines. The firms incur an average of about $1,800
in fees, paid upfront, to take out the credit line. They
pay an average of 8.41 percent plus a risk premium of
39 basis points on any amount drawn from the credit
line and a penalty rate of about 2 percent on any amount
drawn above the stated line amount. To obtain credit
lines, 95 percent of the firms in our sample used col
lateral to secure the line commitment, with about 19
percent using deposits at the bank and 76 percent using
business assets as collateral. The average line com
mitment for our sample firms is a little under $ 1 mil
lion. Over the two-year period covered by our sample,
73
firms on average draw down a little over half of their
credit line.
Empirical strategy
Although we can use our data to look at correlations
between the quantity and price of credit lines and oth
er firm and industry characteristics, in the absence of
further assumptions we can’t be sure whether those
relationships are driven by changes in the supply of
loans or changes in the demand for such loans.
However, one piece of information we observe on
the loans helps us identify the difference between supply
and demand: We see both the amount of the loan asked
for by the firm and the amount granted by the bank.
We argue that if we restrict our analysis to cases where
the amounts asked for and granted are the same, the
resulting differences in prices and quantities across
firms will reflect differences in demand for commit
ment lines rather than supply. You can think of this as
firms submitting an application for a given line com
mitment where the price is posted by the bank. To see
this, consider two firms that happened to demand the
same amount of credit, but differed in some character
istic that led the bank to be less willing to lend to one
firm than to the other. Then we should observe that for
one firm, the amount supplied is equal to the amount
demanded; but for the other, the amount supplied would
be less than the amount initially demanded. Thus, the
differences in the amount (and the price) transacted
would be attributable in that case to differences in fac
tors affecting loan supply. In contrast, by looking at
cases where the amount demanded is equal to the amount
supplied, we can be more confident that any differences
in quantities (and prices) across firms are attributable
to differences in the demand for credit across those firms.
Making this restriction reduces our sample from the
original data set of 1,147 firms to 637 firms. Since no
identification scheme is perfect, we acknowledge that
some of the factors we identify here as contributing to
credit demand may also be contributing to credit supply.
By allowing us to estimate the determinants of
firms’ demand for loans under commitment, this ap
proach also permits us to determine the degree to which
some of the hypotheses about firms’ demand for credit
lines are applicable. To some extent, we can evaluate
the first and third hypotheses—that firms use credit lines
to hedge against deteriorations in their own creditworthiness or to solve problems with informational
asymmetries inherent to other forms of borrowing—
by incorporating risk measures of the firm. It is a bit
difficult in our sample to determine the role of the sec
ond hypothesis—insurance against aggregate declines
in consumer credit. Although our sample period does
cover the aftermath of a recession, the relative tightness
74
TABLE 1
Distribution of firm characteristics
Industry
Percent
Mining and construction
Manufacturing (textile, food, tobacco,
furniture, printing, petroleum)
Manufacturing (rubber, leather, metal,
machinery, equipment, electronics)
Transportation
Trade
Finance, insurance, and real estate
Services (hotels, personal and business
services, auto)
Services (health, legal, engineering)
8
14
19
2
21
24
3
8
State
22
26
7
39
6
Massachusetts
Connecticut
Rhode Island
New York
New Jersey
Notes: The total number of firms in our sample is 637. These
distributions are at account origination.
Source: Panel data set from a large bank.
TABLE 2
Summary statistics
Standard
Variable
Mean
Credit line commitment8
997,274
Utilization13 (two-year average)
51.88
Commitment fee8
1,829
Interest rate on takedown6
8.41
Risk premium spread6
0.39
Overcharge fee spread6
2.01
Net profit growth6
22.48
Net sales growth6
25.32
Total assets growth6
12.91
Risk ratings
5.01
Net cash flow8
178,090
Working capital8
631,034
Years in business
10.03
Total assets8
2,009,239
Number of firms
error
993,012
54.23
331
1.44
0.54
5.11
6.03
2.94
59.34
0.64
131,299
590,953
5.78
1,693,984
637
aDollars.
bPercent.
Source: Authors’ calculations based on panel data set from
a large bank.
of corporate credit during this period is not as great as
it was during the periods studied by other authors. We
can partially test the fourth hypothesis—that firms take
out credit lines for their flexibility—by including proxies
for the firm’s likely need for funds.
2Q/2011, Economic Perspectives
Our main specification is:
(?■ = 30 + P'Przce. + ^NetFundNeeds, +
\VRisk, + ^Collaterals + P54?e,. +
(^Industry, + |:i7 Statet
Q. is the size of the credit line normalized by firm assets;
we do this normalization because credit line demand
may be very different for different sizes of firm.
Price, is a set of contract pricing components, in
cluding fees charged for setting up the line, fees for
overdrawing, the interest rate charged on funds drawn,
and the risk premium spread.
NetFundNeeds.l consists of measures of the mean
and standard deviation of the firm’s net need for exter
nal funds, cash flow, and working capital. Martin and
Santomero’s (1997) model suggests that these param
eters are two important determinants of the size of credit
lines. Since net need for funding is not directly observ
able, we need to proxy for its mean and standard devi
ation. The need for external funds will be greater the
more investment opportunities are available. If firms
are persistently able to find good investment opportu
nities, they will be persistently profitable. Thus, we use
the mean and standard deviation of net profits over our
sample as our proxy for the mean and standard devia
tion of net credit needs. We include cash flow and work
ing capital because externally borrowed funds are needed
less when more internal funds are available.
Risk. is the bank’s risk rating for firm i.
Collateral, consists of two dummy variables—
one for the use of deposits at the bank and one for the
use of business assets. Collateral should matter for two
reasons. First, the posting of collateral helps reduce the
riskiness of the loan to the bank, and thus has some bear
ing on the first and third hypotheses for credit ration
ing. Second, collateral can be considered as one of the
determinants of pricing for the loans. Because collat
eral has this dual role, we break it out separately from
the risk and pricing terms above.
We also control for other firm-level characteristics
that might affect demand for funds. Age. represents the
number of years that firm i has been in business and
the number of years squared. If a younger firm faces
more uncertainty about its growth prospects than an
older firm, it is more likely to commit to a smaller line
and use less of its line commitment. We also include
dummy variables for the firm’s industry (IndustryI) and
the state in which the firm is headquartered (State,).
Although we have argued that we control for one
potential problem—the difficulty in separating supply
from demand—we may still face another problem. It
Federal Reserve Bank of Chicago
may be the case that omitted variables that affect loan
supply happen to be correlated with the regressors,
thereby biasing the coefficients. However, since we
include all the variables observed by the financial in
stitution, we are confident that the errors in the regression
are not related to firm characteristics that might affect
the bank’s supply of loans to the firm. Our approach in
this regard is the same as that taken by Adams, Einav,
and Levin (2009) for auto loans and Karlan and Zinman
(2009) for other consumer loans.
Results
Table 3 presents the model estimates. Firms that
have to pay higher upfront commitment fees, higher
risk premium spreads, or higher usage fees commit to
a smaller credit line, while firms that face a higher penalty
for overdrawing their line commit to a larger credit line.
All of the effects are economically large and statisti
cally significant and jointly suggest that the quantity
demanded is decreasing in the various pricing terms of
the loan—that is, the demand curve slopes downward.
An increase of 1 percent in upfront commitment
fees decreases the line commitment by about 4 percent—
a surprisingly large amount, given the relatively small
average size of the fees. A 1 percentage point increase
in the overcharge fee spread increases the amount of
the credit line by more than 6 percent. Since 1 percent
age point is large relative to the average penalty, but is
well within the 5 percentage points standard deviation
for that variable, normal changes in the spread lead to
very large changes in the size of the credit line. A 1 per
centage point increase in the interest rate—an amount
slightly less than one standard deviation for that vari
able—leads to about a 10 percent decline in the initial
credit line, while an increase in the risk premium spread
of 1 percentage point (about two standard deviations)
reduces the initial credit line by about 18 percent.
Proxies for net funding needs also have a very
large impact on credit line demand. An increase in
average net profit growth, which we would expect to
be positively correlated with future need for funds, of
1 percent raises credit demand by about 16 percent. An
increase of 1 percent in the standard deviation of net
profit growth (which we would similarly expect to be
positively related with the standard deviation of net
funding needs) lowers credit demand by about 15 per
cent. An increase in net cash flow of 1 percent lowers
demand for credit by about 1.75 percent. Although this
result has the right sign (since internal funds should
reduce the net need for funds), its magnitude is small.
Contrary to our expectations, having more working
capital paradoxically raises credit line demand. This
result may arise because working capital may be a
75
predictor of future funding needs.9 The net funding needs
variables, as a group, have a larger effect on credit
demand than any of the other explanatory variables,
suggesting that the fourth hypothesis for what deter
mines demand for loans under commitment—Martin
and Santomero’s (1997) model of firms’ demand for
flexibility in financing—plays an important role.
An increase of 1 point on the risk rating (on an
8-point scale of increasing risk) lowers credit demand
by over 1.5 percent. From Campbell (1978) and Hawkins
(1982), we would have expected that firms fearing
reductions in credit ratings would have demanded more
credit. Our findings here do not support that idea, if we
assume that already riskier firms are more concerned
about deterioration. However, it is possible that rela
tively less risky firms fear credit deterioration more, or
pay relatively higher costs when their credit deteriorates.
The use of collateral, not surprisingly, increases the
demand for credit, more so when collateral is in the form
of deposits rather than in the form of business assets.
We also include, but do not report in the tables,
other measures of firm characteristics that might affect
credit demand. Younger firms hold larger lines of credit,
perhaps because they fear deterioration in creditworthi
ness; each additional year in business increases credit
demand by about 2 percentage points. Firms whose
industry classification places them in the finance, in
surance, and real estate; trade; or service sectors have
larger credit lines than those in mining and construc
tion or manufacturing. There is no substantial variation
in credit line size by state location.
Credit line utilization
Conditional on having chosen the size of the credit
line, firms’ draws on the line should reflect the arrival
of investment opportunities. But when firms must re
peatedly choose lines, line usage should also influence
the timing of such choices and the size of the line. If
firms employ credit lines to give them the flexibility to
take advantage of investment opportunities that would
otherwise disappear, they should take out a new line
before the current one is used up. We frequently ob
serve this in our data: Firms convert the unused portion
of the credit line into a spot loan and take out a new
line of credit.
Since utilization and the size of the credit line may
therefore be jointly determined, we run the same regres
sion as in table 3, replacing the size of the credit line
with utilization (measured as a two-year average of the
total amount drawn by the firm relative to the total
credit line amount). The results, reported in table 4,
are generally in line with expectations and the results
reported in table 3.
76
TABLE 3
Demand for credit lines
Intercept
93.39*’
(39.91)
Price
Log (commitment fee)
-4.02*’
(1-02)
Overcharge fee spread
6.42*’
(2.81)
Interest rate
-10.39*’
(4.09)
Risk premium spread
-17.83*’
(7.37)
Net funding needs
Mean net profit growth
15.88*’
(6.73)
Standard deviation of net profit growth
-14.67*’
(5.93)
Log (net cash flow)
-1.75
(1-21)
Log (working capital)
7.80*
(3.10)
Risk
Risk rating
-1.59*
(0.79)
Collateral
Collateral (deposits)
14.83*
(5.92)
Collateral (business assets)
4.17
(2.63)
Firm characteristics included
Years in business
SIC dummies
State dummies
Yes
Yes
Yes
Adjusted R-squared
0.68
Number of observations
637
‘Denotes statistical significance at a 95% confidence level.
“Denotes statistical significance at a 99% confidence level.
Notes: This table reports the results of an ordinary least squares
regression of credit line amount normalized by firm assets on
measures of price, net funding needs, risk, collateral, age, and
firm characteristics (not reported). Heteroskedasticity-robust
standard errors are in parentheses. The price measures consist
of commitment fees (log thousands of dollars), overcharge fee
spread, interest rate, and risk premium spread (all in percentage
points). Net funding needs are represented by the mean and
standard deviation of net profit growth (percent growth), net cash
flow, and working capital (both log thousands of dollars). Risk
rating is measured on a scale of 1-8, where 8 represents the
highest risk. Collateral is measured by a dummy variable for
each type. All percentage and growth rate figures are expressed
as decimals. SIC indicates standard industrial classification.
Source: Authors’ calculations based on panel data set from a
large bank.
We find that higher upfront commitment fees are
associated with greater usage of credit lines; a 1 percent
increase in such fees raises utilization by about 4 percent.
This may reflect a selection effect: Firms willing to pay
higher fees to establish credit lines may also be in in
dustries in which investment opportunities arise more
2Q/2011, Economic Perspectives
TABLE 4
Usage of credit lines
Intercept
Price
Log (commitment fee)
104.28*’
(32.58)
3.81*’
(1-45)
Overcharge fee spread
2.03*
(1-02)
Interest rate
-4.74*’
(1-18)
Risk premium spread
-7.07*
(3.47)
Net funding needs
Mean net profit growth
Standard deviation of net profit growth
10.57*
(4.72)
-11.42*
(5.61)
Log (net cash flow)
-1.04
(0.69)
Log (working capital)
-1.89*
(0.88)
Risk
Risk rating
Collateral
Collateral (deposits)
Collateral (business assets)
-2.93*
(1-29)
7.19
(5.92)
3.74
(6.93)
Firm characteristics included
Years in business
SIC dummies
State dummies
Yes
Yes
Yes
Adjusted R-squared
0.37
Number of observations
637
‘Denotes statistical significance at a 95% confidence level.
“Denotes statistical significance at a 99% confidence level.
Notes: This table reports the results of an OLS regression of credit
line usage (a two-year average of the percentage of the credit line
used) on measures of price, net funding needs, risk, collateral, age,
and firm characteristics (not reported). Heteroskedasticity-robust
standard errors are in parentheses. The price measures consist of
commitment fees (log thousands of dollars), overcharge fee spread,
interest rate, and risk premium spread (all in percentage points).
Net funding needs are represented by the mean and standard deviation
of net profit growth (percent growth), net cash flow, and working
capital (both log thousands of dollars). Risk rating is measured on
a scale of 1-8, where 8 represents the highest risk. Collateral is
measured by a dummy variable for each type. All percentage and
growth rate figures are expressed as decimals. SIC indicates
standard industrial classification.
Source: Authors’ calculations based on panel data set from a
large bank.
frequently. Overcharge fees have a small but statistically
significant effect on usage. Increases in interest rates and
risk premium spreads lead to lower utilization rates, but
the effects are much smaller than those for credit line size.
The average and standard deviation of net profit
growth affect utilization as expected—the former
Federal Reserve Bank of Chicago
increasing it (by 10 percent for each 1 percentage point
increase); the latter decreasing it (by 11 percent for each
1 percentage point increase). Cash flow and working
capital have negligible effects on usage, possibly be
cause, conditional on having obtained the line, it is less
costly for firms to use external funds (which must be
paid for whether they are used or not) than internal funds.
Riskier firms use smaller amounts of their credit
lines; each one-step increase in risk category decreases
line usage by about 3 percent. This may be consistent
with the hypothesis that riskier firms are reluctant to use
their credit for fear that credit will become more costly
or unavailable if their condition deteriorates further.
Collateral has a large but statistically insignificant
effect on usage. There is no economically or statistically
significant variation in utilization by age of the firm,
industrial classification, or state location.
Conclusion
Firms borrow in order to undertake investment or
to insulate themselves from macroeconomic shocks,
among other reasons. Thus, a better understanding of
firm borrowing not only allows us to better model in
dividual firm behavior, but also may enhance our abili
ty to understand business cycles. Credit lines are an
important source of borrowing, especially for small
firms. There are several competing explanations for
the existence and use of credit lines: hedging against
deterioration in creditworthiness, hedging against aggre
gate reduction in credit, solving informational prob
lems that make it hard for firms to borrow in other
markets, or providing speed and flexibility to enable
firms to take advantage of investment opportunities.
Although a number of researchers have looked at the
determinants of the supply of credit lines, few have
looked at demand; those that have looked at demand
have analyzed publicly traded firms, for which data
are more readily available.
In this article, we look at the demand for credit
lines by privately held firms. Our findings are consis
tent with predictions derivable from several models
of credit line usage. Firms facing higher upfront com
mitment fees, risk premium spreads, or usage fees have
smaller credit lines, while those with higher overdraft
fees have larger ones. Firms with greater profit growth
in the past have larger credit lines, while those with
more internal funds or higher volatility in profit growth
have smaller credit lines. The results for line utilization
are quite similar. We also find that firms rarely exhaust
their credit lines; rather, they convert the unused por
tions of their credit lines to spot loans and take out new
lines. This last finding suggests there is a dynamic in
teraction between line size and usage; it would be of
77
interest to model this relationship in order to develop
new predictions and to link the estimates of firm bor
rowing behavior more directly to models of economic
fluctuations. Finally, although we have tried to separate
the determinants of demand from those of supply, we
have likely not done so perfectly. Thus, some of the
effects we identify may also reflect factors that affect
loan supply.
NOTES
’For further discussion of banks’ roles in solving information
problems in small business lending, see Berger and Udell (1998)
and Petersen and Raj an (1994, 1995).
6For further discussion of the behavior of bank lending during
the financial crisis, see Evans (2008), Bernanke (2009), and Duke
(2009, 2010).
2See, for example, Bernanke and Blinder (1992); Gertler and
Gilchrist (1992); and Kashyap, Stein, and Wilcox (1993).
’See Thakor and Udell (1987); Shockley and Thakor (1997);
Boot, Thakor, and Udell (1987, 1991); Berkovitch and
Greenbaum (1991); Duan and Yoon (1993); and Kanatas (1987).
3From the FFIEC’s Reports of Condition and Income for commer
cial banks. Unused commitments on business credit lines are not
measured directly; the cited figure is derived by taking total unused
commitments and subtracting unused commitments on consumer
credit lines.
4See, for example, Campbell (1978) and Hawkins (1982).
8See Berger and Udell (1992); Sofianos, Wachtel, and Melnik
(1990); Morgan (1994); and Melnik and Plaut (1986b).
9Using other measures of firm growth, such as growth of total
assets, total liabilities, and total sales in our regressions yielded
results that were qualitatively similar.
5See, for example, Blackwell and Santomero (1982); Melnik
and Plaut (1986a); Sofianos, Wachtel, and Melnik (1990); Avery
and Berger (1991); Berger and Udell (1992); and Morgan (1994).
REFERENCES
Adams, W., L. Einav, and J. Levin, 2009, “Liquidity
constraints and imperfect information in subprime
lending,” American Economic Review, Vol. 99, No. 1,
March, pp. 49-84.
Bernanke, B. S., and A. S. Blinder, 1992, “The federal
funds rate and the channel of monetary transmission,”
American Economic Review, Vol. 82, No. 4, September,
pp. 901-921.
Avery, R. B., and A. N. Berger, 1991, “Loan commit
ments and bank risk exposure,” Journal ofBanking
and Finance, Vol. 15, No. 1, February, pp. 173-192.
Blackwell, N. R., and A. M. Santomero, 1982,
“Bank credit rationing and the customer relation,”
Journal ofMonetary Economics, Vol. 9, No. 1,
pp. 121-129.
Berger, A. N., and G. F. Udell, 1998, “The economics
of small business finance: The roles of private equity
and debt markets in the financial growth cycle,” Journal
ofBanking and Finance, Vol. 22, Nos. 6-8, August,
pp. 613-673.
, 1992, “Some evidence on the empirical
significance of credit rationing,” Journal ofPolitical
Economy, Vol. 100, No. 5, October, pp. 1047-1077.
Berkovitch, E., and S. I. Greenbaum, 1991, “The
loan commitment as an optimal financing contract,”
Journal ofFinancial and Quantitative Analysis, Vol. 26,
No. 1, March, pp. 83-95.
Bernanke, B. S., 2009, “On the outlook for the econ
omy and policy,” speech to the Economic Club of
New York, New York, November 16.
78
Boot, A. W. A., A. V. Thakor, and G. F. Udell, 1991,
“Credible commitments, contract enforcement problems
and banks: Intermediation as credibility assurance,”
Journal ofBanking and Finance, Vol. 15, No. 3, June,
pp. 605-632.
, 1987, “Competition, risk neutrality and
loan commitments,” Journal ofBanking and Finance,
Vol. 11, No. 3, September, pp. 449-471.
Campbell, T. S., 1978, “A model of the market for
lines of credit,” Journal ofFinance, Vol. 33, No. 1,
March, pp. 231-244.
Duan, J. C., and S. H. Yoon, 1993, “Loan commit
ments, investment decisions and the signalling equi
librium,” Journal ofBanking and Finance, Vol. 17,
No. 4, June, pp. 645-661.
2Q/2011, Economic Perspectives
Duke, E. A., 2010, “The economic outlook,” speech to
the Economic Forecast Forum, Raleigh, NC, January 4.
__________ , 2009, “A framework for analyzing bank
lending,” speech at the 13th Annual University of
North Carolina Banking Institute, Charlotte, NC,
March 30.
Evans, C. L., 2008, “Discussion on the current econ
omy,” remarks to the Economic Club of Indiana,
Indianapolis, November 21.
Gertler, M., and S. Gilchrist, 1994, “Monetary poli
cy, business cycles, and the behavior of small manu
facturing firms,” Quarterly Journal ofEconomics,
Vol. 109, No. 2, May, pp. 309-40.
Ham, J. C., and A. Melnik, 1987, “Loan demand:
An empirical analysis using micro data,” Review of
Economics and Statistics, Vol. 69, No. 4, November,
pp. 704-709.
Hawkins, G. D., 1982, “An analysis of revolving
credit agreements,” Journal ofFinancial Economics,
Vol. 10, No. 1, March, pp. 59-81.
Kanatas, G., 1987, “Commercial paper, bank reserve
requirements, and the informational role of loan com
mitments,” Journal ofBanking and Finance, Vol. 11,
No. 3, September, pp. 425-448.
Karlan, D., and J. Zinman, 2009, “Observing unob
servables: Identifying information asymmetries with
a consumer credit field experiment,” Econometrica,
Vol. 77, No. 6, November, pp. 1993-2008.
Kashyap, A. K, J. C. Stein, and D. W. Wilcox, 1993,
“Monetary policy and credit conditions: Evidence
from the composition of external finance,” American
Economic Review, Vol. 83, No. 1, March, pp. 78-98.
Melnik, A., and S. Plaut, 1986a, “The economics of
loan commitment contracts: Credit pricing and utiliza
tion,” Journal ofBanking and Finance, Vol. 10, No. 2,
June, pp. 267-280.
__________ , 1986b, “Loan commitment contracts,
terms of lending, and credit allocation,” Journal of
Finance, Vol. 41, No. 2, June, pp. 425-435.
Morgan, D. P., 1994, “Bank credit commitments,
credit rationing, and monetary policy,” Journal of
Money Credit and Banking, Vol. 26, No. 1, February,
pp. 87-101.
Petersen, M. A., and R. G. Rajan, 1995, “The effect
of credit market competition on lending relationships,”
Quarterly Journal ofEconomics, Vol. 110, No. 2, May,
pp. 407-443.
__________ , 1994, “The benefits of lending relation
ships: Evidence from small business data,” Journal of
Finance, Vol. 49, No. 1, March, pp. 3-37.
Shockley, R. L., and A. V. Thakor, 1997, “Bank
loan commitment contracts: Data, theory, and tests,”
Journal ofMoney, Credit and Banking, Vol. 29, No. 4,
November, pp. 517-534.
Sofianos, G., P. Wachtel, and A. Melnik, 1990, “Loan
commitments and monetary policy,” Journal ofBanking
and Finance, Vol. 14, No. 4, October, pp. 677-689.
Thakor, A.V., and G. F. Udell, 1987, “An economic
rationale for the pricing structure of bank loan com
mitments,” Journal ofBanking and Finance, Vol. 11,
No. 2, June, pp. 271-289.
U.S. Census Bureau, 2005, Statistics of U.S. Businesses,
Washington, DC, available at www.census.gov/epcd/
susb/2005/us/US-.HTM.
Martin, J. S., and A. M. Santomero, 1997, “Invest
ment opportunities and corporate demand for lines of
credit,” Journal ofBanking and Finance, Vol. 21, No. 10,
October, pp. 1331-1350.
Federal Reserve Bank of Chicago
79
Fourteenth Annual International Banking Conference
The Role of Central Banks in Financial Stability:
How Has It Changed?
November 10-11,2011
Federal Reserve Bank of Chicago
In conjunction with the European Central Bank, the Federal Reserve Bank of Chicago will
hold its fourteenth annual International Banking Conference on November 10-11, 2011,
at the Bank. The purpose of the annual conference is to address important current issues
affecting international financial markets. This year, we will examine the changing role of
central banks in pursuing financial stability. Prior to the recent financial crisis, there were
significant differences across countries in how and to what extent financial stability was
pursued by central banks. In some countries, the central bank had an explicit stability
objective but did little to actively manage stability other than to ensure liquidity access, out
of fear that more active involvement might distort markets. In other countries, the central
bank prepared formal stability reports and/or pursued financial stability more actively.
Following the global financial crisis, significant reforms have been initiated in many coun
tries to address financial stability more directly, frequently focusing on macroprudential
policy frameworks in which central banks play a more active role.
We are interested in examining a number of important issues associated with the
recent change in emphasis at central banks with regard to financial stability. For example,
what were the cross-country differences in emphasis on financial stability in the past? Did
these differences appear to affect the extent of the adverse impact of the crisis on
individual countries? Can systemic risk be measured and identified? What alternative
macroprudential policy tools have been introduced to address systemic risk? Have views
changed on how to address sources of financial instability, including asset bubbles? What
are perceived to be the major future threats to financial stability? Did the financial sector
grow too big within the pre-crisis financial architecture from a social cost-benefit perspec
tive? Might the pursuit of financial stability have adverse societal welfare implications if
certain financial activities or innovations are limited or prohibited? How potentially effective
might recently introduced reforms be at achieving their stated goals? What major “gaps”
still exist? These and related issues will be addressed at the two-day conference.
As at past conferences, the emphasis of the conference will be on the implications
for public policy. The conference will feature keynote presentations by Janet Yellen, Vice
Chair of the Federal Reserve System; Mario Draghi, Governor of the Banca d’Italia
(invited); and Axel Weber, former President, Deutsche Bundesbank (invited). As usual, the
makeup of the conference will truly be international. Participants from some 35 countries
regularly participate in the conference and include representatives of central banks,
regulatory and supervisory agencies, financial institutions, trade associations, and
academic institutions from around the globe.
We invite you to participate in this important event. Additional information, includ
ing the full agenda and conference and hotel registration details, will be posted on the
conference website as it becomes available:
www.chicagofed.org/lnternationalBankingConference
We hope you can join us in Chicago in November.
Location:
Contact:
Federal Reserve Bank of Chicago
230 South LaSalle Street
Chicago, IL 60604-1413
Ms. Blanca Sepulveda
(312) 322-8340
Blanca.Sepulveda@chi.frb.org
