Full text of Working Papers (Federal Reserve Bank of Richmond) : Did the Financial Reforms of the Early 1990s Fail? A Comparison of Bank Failures and FDIC Losses in the 1986-92 and 2007-13 Periods, Working Paper 15-05

View original document

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

Working Paper Series

Did the Financial Reforms of the Early
1990s Fail? A Comparison of Bank
Failures and FDIC Losses in the 1986-92
and 2007-13 Periods
WP 15-05

Eliana Balla
Federal Reserve Bank of Richmond
Edward Simpson Prescott
Federal Reserve Bank of Richmond
John R. Walter
Federal Reserve Bank of Richmond

This paper can be downloaded without charge from:
http://www.richmondfed.org/publications/

Did the Financial Reforms of the Early 1990s Fail?
A Comparison of Bank Failures and FDIC Losses
in the 1986-92 and 2007-13 Periods
Eliana Balla, Edward Simpson Prescott, John R. Walter
Federal Reserve Bank of Richmond†
May 15, 2015

Federal Reserve Bank of Richmond Working Paper 15-05
Abstract
Two of the most significant banking reforms to come out of the banking problems in the late
1980s and early 1990s were the increase in capital requirements from Basel 1 and the prompt
corrective action (PCA) provisions of the Federal Deposit Insurance Corporation Improvement
Act of 1991 (FDICIA). The PCA provisions require regulators to shut down banks before book
capital becomes negative. We compare failures and FDIC losses on commercial banks in the preFDICIA commercial bank crisis of the mid-1980s to early 1990s with that in the recent financial
crisis. Using a sample of community and mid-sized banks, we find that almost all the same bank
characteristics predict failure and high losses in the two crises. Our results imply that for these
classes of banks, the two crises were very similar. We find that the failure rate in the recent
period was driven more by severe economic conditions than by the increased concentrations in
real estate lending. The analysis suggests that the combination of PCA with higher capital levels
helped reduce failure rates in the recent period. In contrast, the analysis suggests that the reforms
did not help with FDIC losses. FDIC losses on failed commercial banks were approximately
14% of failed bank assets over the 1986-92 period, but increased to approximately 24% over the
2007-13 period. We find that the increased losses are not explained by variations in bank balance
sheets or local economic conditions. Finally, we find that a discretionary accounting variable,
interest accrued but not yet received, is predictive of both failure and higher FDIC losses.
Keywords: bank regulation, bank failures, Prompt Corrective Action, FDIC losses
JEL codes: G21, G28
_____________________
†

We would like to thank Laurel Mazur and John Muth for excellent assistance with this paper. We would also like
to thank Rosalind Bennett, Morgan Rose, Anna-Leigh Stone, Larry Wall, and participants at the 2014 Southern
Finance Association and 2014 Southern Economic Association conferences for helpful comments. The views
expressed in this paper are those of the authors and do not necessarily reflect those of the Federal Reserve Bank of
Richmond or the Federal Reserve System.

1. Introduction
From the mid-1980s to the early 1990s, the United States experienced a severe
commercial banking crisis. There were 998 bank failures between 1986 and 1992 (see Figure 1).
Of non-de novo banks in existence at the end of 1985, those holding 5.4% of bank assets and
6.0% of bank deposits failed and went through Federal Deposit Insurance Corporation (FDIC)
receivership over this same period. 1
Two of the more significant banking reforms that came out of this period were the
increased regulatory capital requirements of Basel 1 and the prompt corrective action (PCA)
provisions in the Federal Deposit Insurance Corporation Improvement Act of 1991 (FDICIA). 2
The latter provisions built on the increased capital requirements of Basel 1 by requiring bank
supervisors to take certain actions against a bank if its regulatory capital drops below certain
thresholds. The purpose of PCA was to force supervisors to intervene in the operations of a bank
and even shut it down, if necessary, before a bank becomes too severely distressed. These
provisions were motivated by the heavy use of forbearance by thrift and bank supervisors in the
1980s and the belief that this forbearance increased the losses to the FDIC, and ultimately
taxpayers, from failed banks and thrifts.
Despite these reforms, the recent financial crisis severely impacted commercial banks.
There were 403 failures over 2007 to 2013 (see Figure 1). Of non-de novo banks in existence as
of the end of 2006, those holding 2.2% of bank assets and 3.1% of deposits went through FDIC
receivership between 2007 and 2013. 3
Of banks that failed, FDIC losses were large in both periods but significantly larger in the
later period. Over 1986-92, the FDIC’s losses on failed banks were nearly 14% of these banks’
assets net of book equity. 4 Yet, despite the implementation of PCA, the FDIC’s losses on failed
banks over the period 2007-13 were significantly higher, approximately 24% of these banks’
assets net of book equity.
1

These fractions were calculated by using as our denominator total commercial bank assets at the beginning of the
period and then using as the numerator the assets or deposits of a failed commercial bank at the beginning of the
period.
2
Basel I and FDICIA were implemented in the early 1990s after the bank failures in our 1986-92 sample.
3
Citibank received assistance in the form of loss protection in 2008, while similar protection was developed for
Bank of America in early 2009 but never implemented. Neither of these large banks went through receivership, so
they are not included in the numerator. For this reason, if we just consider the smaller banks in our sample, the
numbers would be higher.
4
We adjust assets by subtracting book equity at time of failure. This adjustment takes into account the loss
absorption capacity provided from book equity -- if there is any -- at time of failure. In the paper, we will typically
refer to this as the adjusted loss ratio.

This paper compares commercial bank failures and FDIC losses on failed banks over
these two crises. We use a sample of established community and mid-sized banks. We exclude
large banks and de novo banks because they have different characteristics than other banks and,
in the case of large banks, some of them received substantial government assistance that
prevented their failure.
Our approach is to use bank characteristics in 1985 and 2006, right before each crisis
fully developed, to estimate failure probabilities and FDIC losses over each period. By doing
this, our estimation strategy can be seen as analyzing how banks, as defined by their
characteristics, respond to a set of severe financial shocks. We estimate a regression model for
both periods that identifies bank factors and state-level economic conditions that are related to
bank failure and the size of FDIC losses. We use a Heckman selection model because, in
addition to being interested in failure probability estimates in each period, we want to account for
selection effects when modeling FDIC losses from failed banks. We then compare the two
periods.
One of the biggest differences between the two periods is the increased concentration in
commercial real estate lending and construction and land development lending (CLD) among
community banks in the recent period. We analyze the degree to which these and other changes
to bank characteristics account for differences in failure rates and FDIC losses by running several
counterfactual experiments. In one, we take the characteristics of banks in 1985 Q4 and evaluate
failure probabilities and FDIC losses conditional on failure using the estimated model for the
2007-13 period. In another, we do the converse.
We find that failure probabilities are most influenced by macroeconomic conditions.
Banks with the characteristics of those in 1985 Q4 would have failed at a much higher rate if
subject to the state-level economic shocks of 2007-13. Similarly, banks with the characteristics
of those in 2006 Q4 would have failed at a much lower rate if subject to the state-level economic
shocks of 1986-92. There are effects on bank failures from the changes in bank balance sheets,
but these effects are much smaller. We find that the higher capital levels of banks in 2006 Q4
offset the increased risks from concentrated real estate lending. The analysis suggests that the
combination of PCA with higher capital levels helped reduce the failure rate. It also suggests that
community and mid-size banks were not necessarily excessively risky going into this crisis.
In contrast, we find very little effect of changes in bank characteristics or state-level
economic conditions on the size of FDIC losses. In our counterfactual exercise, losses would
3

have been large in the 2007-13 period even if banks had not been so concentrated in CLD
lending and looked like they did in 1985.
While our interpretation of the counterfactual exercise is that PCA is not directly related
to the size of the losses, the analysis also shows that PCA was ineffective along this dimension.
One purpose of PCA was to shut down a failing bank before its losses got too big, and on this
dimension it failed. We argue that PCA was doomed to fail because of two interacting factors: 1)
When a bank fails, the market value of its assets is significantly less than its book value; 2) PCA
triggers were set at levels such that capital levels of a bank on the path to failure were only a few
hundred basis points higher than pre-PCA. We show that, on average, banks in the recent period
were put into receivership while their capital was still positive, as PCA requires. However, given
that in this sample the market value of a failed bank’s assets are typically anywhere between
75% and 85% of book value during a crisis, a failed bank simply does not have enough capital to
absorb losses without calling on the deposit insurance fund.
A necessary condition for these losses to be so high, when book values are not negative,
is that the book accounting values of a failed bank’s assets dramatically exceed their economic
value. Along these lines, we find that an accrual accounting variable, interest accrued but not yet
received, significantly predicts bank failure and FDIC losses in both periods. 5
One of our striking findings is that the two crises have very similar qualitative effects.
Virtually the same variables predict failures in both periods, though the sizes of the estimates
differ. We find that construction and land development (CLD) lending increases the failure
probability in both periods, but it is insignificant for losses. Commercial and industrial (C&I)
lending has the same effect. Bank size lowers losses in both periods, while it lowers the failure
probability in the later period. Capital reduces the failure probability in both periods. Securities
holdings also lower failure probabilities, while also lowering losses in the later period. Not
surprisingly, core deposits lower failure probabilities and lower losses in the early period.
Economic conditions have the expected effects. We find that state real estate conditions and
increases in unemployment predict failure in both periods. Similarly, non-performing loans and
non-CLD commercial real estate lending predict failure in both periods. Residential lending is
only significant in the earlier period, lowering the failure probability.
One aspect of our analysis should be kept in mind when assessing PCA and the higher
capital requirements. The model is not a structural model, so the estimates also pick up other
5

Bovenzi and Murton (1988), James (1991), and Osterberg and Thomson (1995) looked at this variable using
samples that overlap with our 1986-92 period, but they only examined FDIC losses and not bank failures.

regulatory factors and governmental actions. In particular, there were large government
interventions in the recent crisis, such as the Troubled Asset Relief Program, the Small Business
Lending Fund, and the expansion of deposit insurance, which could have had effects on bank
failure probabilities and FDIC losses. Nevertheless, the analysis allows us to conclude that even
with the help of these additional interventions, PCA did not succeed in lowering FDIC losses and
that the same bank characteristics predict failure in both periods.

2. Literature Review
Our strategy is to use a Heckman selection procedure (Heckman, 1979) to jointly
estimate the probability of bank failure and the losses to the FDIC conditional on failure. The
empirical literature has typically looked at these separately.
Much of the literature that looks at FDIC losses on failed banks is based on data samples
from the 1980s and early 1990s. Using a sample of failed banks from 1985 and 1986, Bovenzi
and Murton (1988) regressed the losses on measures of asset quality as well as a few other
variables right before bank failure. James (1991) built on this analysis by using a larger sample
and including additional variables like the book value of equity and core deposits. Osterberg and
Thomson (1995) did a similar analysis to James (1991) but used Call Report data and a sample
from 1984 through 1992. They also regressed losses on bank data at various lags prior to failure.
More recently, Bennett and Unal (2014) examined FDIC losses over the 1986 to 2007
time period. The question they are interested in is the effect of the type of FDIC resolution on
FDIC losses. On average, FDIC losses on private-sector reorganizations are less than those on
failed banks that it liquidates. However, once they control for selection bias, they find that during
periods of industry distress private-sector reorganizations of a failed bank are costlier than
liquidation, while during normal time periods this result is reversed.
The second relevant literature is on the causes of the bank failure. Most of the research on
the banking troubles of the 1980s and early 1990s found that commercial real estate
concentrations played a significant role in failure. Fenn and Cole (2008) found this to be the case
for bank failures from 1986 to 1992, and they found that construction loans played a larger role
in bank failures than permanent loans. Cole and Gunther (1995) looked at the likelihood and
timing of bank failure for banks that failed over the 1986 to 1992 period. They found that
commercial real estate concentrations increased the likelihood of bank failure but are unrelated
to bank survival time. Whalen (1991) estimated a proportional hazards model of bank failure
5

and, unlike the other papers, found that commercial real estate was insignificant, though this may
be because all types of commercial real estate lending were lumped together in that analysis.
Less work has been done on the causes of bank failures in the recent crisis. Cole and
White (2012) analyze Call Report data from 2004 to 2008 in order to determine the factors that
led to bank failures in 2009. They find that commercial real estate lending, particularly in the
area of construction and development, is a strong early predictor of bank failure. Noting that this
result is consistent with research on earlier bank failures, the authors stress the importance of
differentiating between commercial and residential real estate when evaluating a bank’s portfolio
and using these data in conjunction with CAMELS ratings to assess commercial bank risk. GAO
(2013) also emphasizes the role of commercial real estate loans, construction and land
development (CLD) loans in particular, as a cause of bank failures in the recent crisis. A paper
that looks at geographic factors is Aubuchon and Wheelock (2010). They find that failure was
also connected to regions with distress in real estate markets and declines in economic activity.
Jin, Kanagaretnam, and Lobo (2011) examine the role of auditing quality in predicting bank
failure. They find that a bank audited by reputable auditors has a lower probability of failure.
This finding is in line with our interest in accounting lags in the recognition of true bank
conditions.
Our paper builds on these two literatures – one looking at reasons for failure and the
second looking at FDIC losses – by jointly estimating a model of failure and losses, conditional
on failure. A model that considers both kinds of effects is needed to evaluate the effectiveness of
PCA.

3. The Mechanics of Bank Failure, Prompt Corrective Action, and FDIC Losses in the Two
Periods
Banks are not subject to the bankruptcy code. Instead, when a bank becomes severely
distressed, it can be put into receivership by its chartering agency, which will either be the Office
of the Comptroller of the Currency for nationally chartered banks or its state regulator for statechartered banks. Once a bank is put into receivership, the FDIC handles its disposition. Before
FDICIA, the FDIC could resolve a bank in any way it chose as long as it was less costly than a
deposit payoff, which is basically a liquidation of the bank in which insured depositors are made
whole by sales of the bank’s assets with any shortfall being covered by the FDIC. Since 1991,

the FDIC has been required to resolve the bank in a way that is the least costly to the deposit
insurance fund.
In the past, chartering agencies had some flexibility as to when they put a bank into
receivership, and this flexibility was used at times in the 1980s to practice forbearance, that is, to
keep insolvent banks operating (White, 1991). In response, FDICIA required regulators to follow
PCA, under which a bank faces restrictions on activities when its capital drops below certain
levels. A bank that is well capitalized does not face any restrictions. A bank is considered well
capitalized if its risk-based capital ratio is 10% or more, its Tier 1 risk-based capital ratio is 6%
or more, and if its leverage ratio is 5% or more. As these capital ratios drop below various
triggers, a bank can become undercapitalized, significantly undercapitalized, and critically
undercapitalized, the latter being when its ratio of tangible equity to total assets is 2% or less. At
various levels of undercapitalization there are restrictions on a bank’s activities, such as
restrictions on paying dividends, limits on growth and funding, and limits on bonuses paid to
senior executives. When a bank is critically undercapitalized, the bank must be put into
receivership or conservatorship within 90 days. 6
Once a bank is in the hands of the FDIC, the FDIC has several means of disposing of it,
though since 1991 the FDIC has been required to do so in a way that is the least costly to the
deposit insurance fund. When the FDIC disposes of a bank, it can either keep it in the private
sector or liquidate it. In the former, this can be done by selling the whole bank or doing a
purchase and acquisition agreement in which part or most of the bank is sold, usually at a
negative price. 7 If a bank is liquidated, then insured depositors are paid off and the receivership
manages the assets in a way that maximizes recoveries that are paid out to the bank’s claimants,
including the FDIC. For more details on how bank failures are resolved, see FDIC (1998) and
Bennett and Unal (2008).
The most common type of transaction, particularly during the recent crisis, is a purchase
and acquisition (P&A) transaction. In this kind of transaction, the acquiring bank assumes either
all or some of the failed bank’s liabilities and purchases all or some of the failed bank’s assets.
Any assets that are left after a P&A transaction are managed and then sold over time by the
receivership. One feature of many P&A transactions, particularly in the first few years of the

See Spong (2000) for a description of PCA.
The FDIC can also provide open-bank assistance, which keeps the existing bank operating. Since 1992 this has
only been used for the “ring fencing,” that is, loss protection, provided to Citibank in 2008 and offered, but never
implemented, for Bank of America in early 2009.

recent crisis, is the use by the FDIC of loss share agreements. These agreements leave a set of
assets in the hands of the acquiring bank and the FDIC takes on or shares in losses on these
assets that exceed some threshold.
The cost to the FDIC is essentially the negative of the market value of a bank (see
Bennett and Unal (2008) for more details) and the market value of a failed bank is

Market value of assets – deposit liabilities + franchise value – receivership costs,

where the franchise value represents the value to an acquiring bank of intangibles like core
deposits or a branch network.
The loss reported depends on what the FDIC can sell a bank for as well as how much is
paid to depositors. In all of these transactions, the FDIC paid off insured depositors in whole. 8
What a bank is willing to pay for a failed bank or part of a failed bank depends on a lot of
factors, including the quality of the assets, the value of the bank’s charter, its core deposits, and
the loss-sharing agreement, if one exists. The FDIC takes these numbers and adds, according to
some rule, its costs from closing the bank. This gives the reported loss numbers.
On average, the banks that failed during the recent crisis had non-negative equity capital
when they were put into receivership, as PCA required. Nevertheless, the losses to the FDIC
were enormous, which means that the market value of each bank’s assets had to be significantly
less than its book value.
Table 1 reports FDIC losses on commercial banks expressed as a percentage of assets net
of book equity. Losses are high regardless of the time period, but they are significantly higher in
the 2007-13 period, despite occurring under the PCA regime. In the 2007-13 period, weighted
losses are 24% while un-weighted losses are 30%. There is clearly a size effect as losses decline
if observations are weighted by assets of the failed bank. Similarly, there is a de novo effect in
that these are much more expensive to resolve, but they are a relatively small share of the
number of failed banks and an even smaller share of failed bank assets, so they do not materially
impact the totals.

In the recent crisis, virtually all depositors were insured. Before September 2008, deposits were insured up to
$100,000 dollars. In September 2008, during the financial crisis, the FDIC extended deposit insurance to up to
$250,000 and for a period its Temporary Liquidity Guarantee Program provided full coverage to all non-interest
bearing deposit transaction accounts.

There are also differences in the distribution of losses between the two periods. Figure 2
shows these distributions. In the 1986-92 period, there is a substantial fraction of banks for
which the losses are under 10% of assets. This is not true in the later period where the
distribution of losses looks more symmetric. In both periods, however, there are some banks with
losses that exceed 50% of assets.
PCA relies on book capital triggers to determine when to shut down a bank. Figure 3
reports the average capital ratio for failed banks in the 16 quarters prior to failure. In the 1986-92
period, the average capital level of a bank in the quarter before failure is about -1.5%. It was this
kind of observation, along with the high losses, that contributed to devising the PCA provisions.
In contrast, the average capital of failed banks in the 2007-13 period was positive, about 1.5%, in
the quarter before failure. The PCA critically undercapitalized level is 2%, so supervisors
faithfully carried out this PCA provision. However, as our analysis will show, given the size of
bank losses that were experienced, having an extra 3% of equity capital at time of failure did not
provide much of an extra buffer to absorb losses.

4. Changes in Bank Activities
One striking difference between the two periods is the increase in CRE and CLD lending
by banks in our sample of community and mid-sized banks. Tables 2a and 2b show just how
dramatic these changes were. For each period, Table 2a reports bank assets expressed as a
percentage of assets for banks that failed. Non-farm, non-residential real estate (CRE) increased
from 6% to 21% of bank balance sheets, while CLD lending increased from 4% to 22%.
Conversely, consumer loans dropped from 13% to 2%, and commercial and industrial loans
declined from 19% to 11%.
Table 2b reports asset concentrations for all banks in our sample, those that failed and
those that did not. A similar qualitative pattern is observed to that of failed banks, but the
quantitative changes are much smaller. For example, CLD lending only increased from 2% to
7% of assets.
The literature has found that commercial real estate and CLD lending, in particular,
increase the chance of bank failure, so it is possible that the increase in losses between the two
periods was driven by this change in bank characteristics. In the following analysis, we will use
our statistical models to evaluate this conjecture.

5. Data and Sample Construction
We examine commercial bank failures in the periods 1986-92 and 2007-13. 9 There were
a total of 998 and 403 failures in those periods, respectively. In our main specifications, we
model a subsample of 713 and 306 failures in each respective period because we exclude large
banks and banks that were in de novo status in 1985 Q4 and 2006 Q4 or were started during the
sample periods.
We set a threshold for large banks in the later period at $50 billion of assets. We chose
this level because it is the threshold at which regulators will consider a bank systemically
important. For the earlier period, we deflate this threshold by the growth in bank assets from
1985 Q4 to 2006 Q4 to get a threshold of $14.8 billion of assets. 10
With general agreement among analysts that the crisis began in the second quarter of
2007, the 2006 Q4 date is a natural date to use before the start of the recent financial crisis. For
the previous banking crisis, we use 1985 Q4 as our starting point because the FDIC does not
report losses prior to 1986. 11 We use quarterly Reports of Condition and Income (Call Report)
that banks file with their regulators. 12 These regulatory reports provide detailed information on
the size, capital structure, and asset and liability composition of each commercial bank. We
merge this sample with the FDIC’s Historical Statistics on Banking (HSOB) dataset, which
includes information on the dollar amount of estimated losses to the FDIC from bank failures. 13
In this paper, all loss estimates to the FDIC are as of December 31, 2013. 14 Note that, as is the
case with most papers in the literature, we use loss estimates. The FDIC provides an estimate of
losses that they update as contractual agreements like loss-share agreements on purchases and
acquisitions or asset dispositions are completed, so there is a possibility that the loss data will
change. 15 Driven by the importance of real estate lending for our empirical setup, we use

We do not consider failure of savings and loans, savings banks, or credit unions.
Only one failed bank is excluded because of these thresholds. Republic Bank of Dallas had $15.8 billion of assets
in 1985 Q4.
11
There were numerous bank failures before 1986, but there were fewer in those years than in 1986. The year with
the most number of failures is 1988.
12
As reported to the Federal Reserve Board by the FFIEC 031 and 041 reporting forms.
13
We filter the FDIC’s historical loss data by “charter type” commercial banks. We examine failures within the
United States proper only (i.e., excluding Puerto Rico and other territories).
14
The FDIC updates losses on an annual schedule, in December of each year. The Failures and Assistance
Transactions database is updated as needed, with the most recent update occurring in December 2014. The data for
this paper was gathered in September 2014.
15
Bennett and Unal (2014) report that as of April 10, 2014, the receiverships for only 21 of the 510 banks that failed
since 2007 have been terminated.
10

Corelogic’s Home Price Index (via Haver Analytics) to create a measure of real estate conditions
during each period. 16
Our dependent variable (“adjusted loss ratio”) is the ratio of the cost to the FDIC of a
given bank’s failure divided by that bank’s consolidated assets at the time of failure, net of its
book equity. 17, 18 Following the literature and applying knowledge from bank supervisory
practices, we select a number of balance sheet and income statement variables to reflect the
business model and financial condition of each institution in our sample.
Our baseline regressions are based on a Heckman selection model with bank-specific
financial ratios measured at the beginning of the period, so failure probability and FDIC losses
occurring any time within the period are explained by financial fundamentals measured in 1985
Q4 and 2006 Q4 for the early and later period, respectively. As described in the introduction, de
novo banks have different fundamentals and resolution costs so we exclude them from the main
analysis. We conduct our main regression analysis on two samples of established banks that
were in existence at the beginning of each sample period. For the early period, we take
established banks that filed a Call Report in 1985 Q4. For the second period, we take established
banks that filed a Call Report in 2006 Q4. These criteria result in a regression sample of 12,658
at the end of 1985, of which 713 failed by 1992. The later period regression sample includes
6,590 banks at the end of 2006, of which 306 failed by the end of 2013. Exact definitions and
sources of the variables are included in Table 3. Summary statistics of the samples used in these
baseline regressions are reported in Table 4.
The bank Size variable is represented as the natural logarithm of a bank’s assets,
measured in thousands. The Capital Ratio variable reflects shareholder’s equity and is
normalized by assets. The Securities Ratio is created by adding the book value of all held-tomaturity securities for the early period (normalized by assets). Because the fair value of all
available-for-sale securities is reported since 1992 only, the later period reflects this addition.
The Earnings Ratio is created by dividing net income by total assets.
We use a number of real estate lending measures in this study. The Construction and
Land Development Loan Ratio is total construction and land development loans divided by total
16

We match the HPI data to each respective bank by the state in which the bank is headquartered.
In the rare cases where the last reported Call Report quarter and the FDIC listed failure date do not correspond, we
drop these observations from the dataset. It is not clear why this discrepancy occurs, but we opted to drop the banks
in which it occurred for consistency.
18
Strictly speaking, the bank’s asset value is measured at the end of the quarter prior to failure, since no Call Report
is filed in the quarter in which it failed.
17

loans. The 1-4 Family Real Estate Loan Ratio is represented by real estate loans secured by 1-4
family residential properties divided by total loans. With CLD and 1-4 family residential lending
measured separately, the Other Real Estate Ratio captures remaining commercial real estate
loans and is calculated by subtracting the ratios for construction and land development loans and
loans for 1-4 residential properties from the overall real estate loan ratio. 19
The Commercial and Industrial Loan Ratio is measured as the sum of commercial and
industrial loans to both U.S. and non-U.S. addressees divided by total loans. The Agricultural
Loans Ratio is represented by loans to finance agricultural production divided by total loans.
The Consumer Loan Ratio is constructed differently in both periods due to differences in Call
Reports. In the early period, it is represented by the sum of credit cards and related plans and
other loans. In the later period, the ratio is measured by the sum of loans to individuals for
household, family, and other personal expenditures, as well as credit cards, other revolving credit
plans, and other loans. We also construct ratios that measure a bank’s asset quality. The NonPerforming Loans Ratio is calculated by adding loans that are 90+ Days Past Due to NonAccrual Loans and dividing by total assets.
Turning to liabilities, the Core Deposit Ratio is measured differently in both periods. In
the early period, the variable was constructed by dividing the sum of total transaction accounts,
money market deposit accounts, and total time deposits less than $100,000 divided by total
deposits. The later period is the same variable specification with the exception of the addition of
other non-transaction savings deposits (excluding money market deposit accounts). The FHLB
Advance Ratio is only reported for the later period and measures the sum of FHLB advances with
a remaining maturity or next re-pricing date of one year or less, one to three years, and over five
years divided by total liabilities and minority interest.
We use two variables that capture loan accounting discretion on the part of bank
management: interest receivables and loan loss reserves. The first variable is a balance sheet
measure of interest income that is accrued but not yet collected. In the early period, the Call
Report reports income earned but not yet collected on loans. In the later period, the Call Report
reports interest income accrued or earned but not yet collected on earning assets. 20 For our
19

The main components of other real estate are multi-family real estate loans, which are secured by multi-family
residential properties divided by total loans, and non-family/non-residential loans, which cover properties like
hotels, churches, hospitals, golf courses, and recreational facilities. (This excludes loans for property and land
development purposes, which mature in 60 months or less.)
20
Accrued interest receivable related to securitized credit cards is not included in the later period definition of this
variable.

purposes, we take these assets and divide them by total assets to obtain what we call the Interest
Receivable Ratio. The second accounting variable is the Loan Loss Reserve Ratio, which is
defined as the allowance for loans and leases divided by total loans and leases, net of unearned
income in the early period and the sum of the allowance for loans and leases and allocated
transfer risk reserves in the later period.
We use two variables that measure local economic performance. The first is Peak to
Trough, which is a constructed measure of deteriorating real estate conditions in the state in
which each bank is headquartered. It is calculated by subtracting the state minimum from the
state maximum Corelogic Home Price Index value, as reported by Haver Analytics, and dividing
by the minimum. This calculation is only made for states where the maximum point came before
the minimum. It is coded with a value of zero for states that experienced no decline in house
prices. This variable measures whether there was a drop in house prices in any sub-period of
each period and, if so, what the largest such percentage drop was. This measure is particularly
useful in the earlier period because house prices declined in 12 states, located primarily in the
East Coast and oil-producing regions, but they did not decline at the same time. For the later
period, 2006 house prices were the peak or close to the peak for most states and then dropped
significantly for all states except North Dakota.
The second variable that we use to measure local economic performance is
Unemployment Increase, which measures deterioration in labor markets in the state in which
each bank is headquartered. In particular, we take the difference between the maximal and
minimal unemployment rate over the period at the state level, conditional on the maximal
unemployment occurring after the beginning of the period. If a state’s unemployment level only
decreased over the period then it is coded as a zero.

6. Regression Results and Analysis
We estimate the same model across the two periods. This increases the chance of omitted
variables in the analysis. Nevertheless, we adopted this strategy in order to compare the periods
closely and to allow for the counterfactual exercises.
Given the potential correlation between failure and losses given failure, we use a
Heckman selection model as our empirical strategy. The main regression results shown in Table
5 are from a two-step estimator. The selection problem arises because losses to the FDIC are
observed only for banks that fail. Banks that fail may differ in important unmeasured ways from
13

healthy banks and in ways in which error terms in failure probabilities are correlated with error
terms in losses. The selection stage has a binary dependent variable taking the form of failure=1
and 0 otherwise within the period. In the outcome stage, we model the size of the losses for
failed banks.
We measure the financial condition of each bank at the beginning of the period studied
(1985 Q4 and 2006 Q4, respectively). There are two advantages to this approach. First, a bank’s
characteristics can change leading up to failure as it sells assets and changes its funding profile.
Second, the Heckman approach lends itself naturally to estimation of a cross-section selection
model. The disadvantage to this approach is that condition is often measured several years before
failure and time to failure varies across banks. Ideally, our estimation would occur in a crosssectional time-series setting. However, there are very few observations of failure-quarters
relative to quarterly observations for all banks in each period for this to be viable. 21 We pool
failures in each period and model the two samples separately. We use a limited set of
explanatory variables in order to maintain consistency across the two periods, to be able to
compare the effects, and to better serve the counterfactual exercises in the next section. In some
cases, notably the use of FHLB Advances, we are limited because of differences in reporting
between the two periods. 22
For our selection equation, we use all the variables previously described. These can be
broken into several types. There are variables that describe the economic conditions in the bank’s
state, the business model of the bank in terms of lending and funding, performance ratios, and
variables that capture the role of accounting measures.
We use a more parsimonious model of determinants of losses to the FDIC. We focus on
size, security holdings, CLD lending, C&I lending, core deposits, and the interest receivable
variable. A potential limitation of the loss-ratio regression is that we are not measuring
determinants of the demand for failed bank assets and deposits, nor are we controlling for type of
resolution (as is the focus of Bennett and Unal (2014)). However, by including securities, core
deposits, and size, we are picking up proxies for the franchise value of a bank. Securities are
liquid and thus easier to sell, core deposits are valuable to acquirers because of their stability, and
larger banks usually have larger branch networks. We also include two types of lending that have
21

We were not able to achieve convergence in Stata when using panel Heckman estimation.
When we include FHLB Advances in our robustness work, we find that in the later period this source of wholesale
funding is not statistically significant for predicting failure. When a bank is put in receivership, the FDIC pays the
FHLB in whole, so we also added them to the loss equation. Again, the variable was not significant. For this reason,
these results are not reported.
22

been shown elsewhere in the literature to be associated with asset quality problems, commercial
and industrial lending, and commercial real estate lending (including its components).
We exclude a variety of variables from the loss equation. However, the excluded variable
that we consider for our exclusion restriction is the Capital Ratio. While it is certainly correlated
with failure probability because regulators shut a bank down when this ratio gets low, it should
not necessarily be correlated with losses. Once a bank’s capital is exhausted, it is insolvent, and
if a bank gets to that point, how much capital it had a few years before should be irrelevant for
losses. Indeed, in most specifications we examined, this variable was insignificant if added to the
loss equation.
We use two variables that capture the severity of economic conditions. The first is our
Peak to Trough variable, which measures declines in house prices. The second is Unemployment
Increase, which measures increases in the unemployment rate. Both are measured at the state
level and associated with a bank by the state where the bank is headquartered. We find that the
coefficient on each is positively correlated with bank failure.
We include three real estate lending variables. Based on historical experience, 1-4 family
residential lending was treated by banks and their supervisors as a safe type of lending entering
the financial crisis. Consistent with this view, in the earlier period, its coefficient on the selection
regression equation is negative and significant. In the later period, it is positive but not
statistically significant. CLD Loan Ratio enters both the selection and outcome stages and has
differing effects across the two stages. In both periods, higher CLD concentration is positively
associated with failure and is statistically significant. It is not, however, statistically significant in
the loss equations and the sign is even negative in the earlier period. The Other Real Estate
concentration ratio is also positively associated with failure and is statistically significant.
The C&I Loan Ratio looks like the CLD Loan Ratio in that it is statistically significant
and positively associated with failure. While positive, it is insignificant, however, for the loss
equations. In contrast, Securities are often considered safe assets and, consistent with this
perspective, we find that the larger the share of assets held in the form of securities, the lower the
probability of failure in both periods. The coefficient in the loss regression is only statistically
significant in the later period (although note that the fair value of available-for-sale securities is
only used in the definition of this variable in the later period).
Higher core deposits reduce failure probabilities in both periods and reduce losses in
both periods, though the coefficient on the loss equation is only significant in the later period. A
15

higher Capital Ratio is, unsurprisingly, related to a lower probability of failure. However, the
Loan Loss Reserve variable, while negatively associated with failure, is not statistically
significant. For Size, we find it is negatively associated with failure but only statistically
significant in the earlier period. More interesting, however, is that it is negatively associated with
losses and statistically significant. It is possible that the larger bank networks provide some
franchise value to acquirers.
Results for measures of bank productivity are as expected. The Earnings Ratio is
negatively associated with failure and statistically significant. Similarly, a higher NonPerforming Loan ratio is positively associated with failure and is statistically significant in both
periods.
A variable that is highly predictive of failure and losses is the Interest Receivable Ratio.
Reported as “Interest earned but not collected” in Call Reports of the earlier period, this variable
refers to interest payments due for loans (in all loan categories) that are accruing interest. 23 The
size of this variable reflects two factors. The first factor is payment structure. For example, one
loan could be due on 7/15 and another on 7/31. Different due dates would then result in different
reported accrued interest for these two loans even if the only way in which they differed was the
payment date. Furthermore, some commercial loans have an extended payment period (quarterly
or twice a year). All else equal, a bigger share of these loans would result in a bigger interest
earned but not collected variable.
The second factor is accounting discretion. A bank has some leeway in its accounting
treatment of loans, so an asset that is not being repaid could be treated as accruing income when
its probability of default has gone up. In this case, the size of the Interest Receivable Ratio
reflects loans that, in the future, would probably end up in non-accrual status. When a loan
moves to non-accrual status, the interest should be backed out of the “interest earned but not
collected” asset so this variable will mechanically drop in value in that quarter’s Call Report
filing. It is possible then that this variable would be correlated with future non-accrual loans.
In separate analyses, we examined the correlation between the lagged values of the
Interest Receivable Ratio and loans that are Non-Accrual and loans that are 90 Days Past Due,
the two components of our Non-Performing Loan Ratio. In the 1986-92 period, we found that
there was a positive correlation of 0.21 between Non-Accrual and Interest Receivable Ratio
lagged at eight quarters. This correlation stays positive but declines with the number of lags and
23

Some of the literature on FDIC losses has identified the importance of this variable (Bovenzi and Merton, 1988;
James, 1991), but to our knowledge its connection to bank failure has not been previously identified.

is 0.12 for the contemporaneous correlation. The correlation between 90 Days Past Due and
Interest Receivable Ratio is 0.20 at eight quarters and increases to 0.28 for contemporaneous
correlation. Interestingly, correlations are different in the later period. The correlation is around
-0.10 with Non-Accrual for all the lags, while it is about 0.15 with 90 Days Past Due for all the
lags. It seems that there are different dynamics with this variable in the two periods even though
it is highly significant in the regression model for both periods.
Although the economic significance of the coefficients cannot be interpreted directly
from Table 5, statistical significance and the direction of the effects can be compared across the
two periods. To summarize, the results reported in Table 5 are directionally in line with
expectations and qualitatively similar across the two periods for several regressors. It is striking
that, although we are measuring the structure of the balance sheet and the performance of the
bank long before failure for a large portion of the failed banks, we still find strong statistical
significance. Management’s decisions on the structure of the balance sheet made long before the
crisis have significant influence on outcomes during the crises.
In Table 6, we report the size of the selection effects by comparing the Heckman loss
equation with an OLS-estimated loss equation. Note that the Heckman correction serves to shift
the conditional expectations of those banks likelier to fail due to unobservable factors in the right
direction. The reported negative lambda (which is the coefficient on the inverse Mills ratio)
implies that the unobservables of the selection stage and the unobservables of the outcome stage
are negatively related. This is consistent with smaller coefficients in the Heckman outcome stage
compared to regular OLS. Note that the second period coefficients are not significant for
standard loss determinants as compared to OLS, suggesting that in regular OLS analysis (that
does not correct for the conditional nature of observing FDIC losses) several regressors are
picking up the indirect effect on losses through the probability of observing those losses given
failure.
To interpret the economic significance of the coefficients, we calculate marginal effects.
The selection equation is a probit equation, which is nonlinear, so marginal effects cannot be
directly computed from the coefficients. Instead, we follow the literature and evaluate marginal
effects at the mean values of the independent variables. However, for failure rates, rather than
reporting the marginal effect from an infinitesimal change in a variable, we calculate the effect of
increasing each independent variable, holding the other variables at their means, by one standard
deviation.
17

Table 7 reports these effects. First, note that in the first row the failure probabilities
evaluated at the mean, in both periods, are slightly more than 1%, which is much less than what
is observed in the data. The reason for this is that failure probabilities are much higher for banks
with large deviations from the mean for certain variables.
The calculated marginal effects can be substantial at the mean values. In the earlier
period, the largest effect is increasing the Interest Receivable variable by one standard deviation.
The house price index also has a substantial effect. Both increase the failure probability by over
100 basis points. Also important is capital, which reduces the failure probability by 82 basis
points. In the later period, the largest effects are from the real estate variables. The CLD Loan
variable has a particularly big effect, raising failure rates by 412 basis points, while Other Real
Estate raises it by 176 basis points.
For the loss equation, the reported marginal effects are conditional on the loss
observation taking a positive value, with values of the remaining explanatory variables held
constant at their means. If a variable enters both estimation stages, that variable’s combined
effect is reported; the direct effect to the loss variables and the indirect effect through selection.
This is because a change in an independent variable not only affects the size of the loss ratio but
also the probability that an observation is in the sample. Here, we provide a different unit of
measurement than for the failure probability. In particular, we calculate the effect of the marginal
change by multiplying the marginal change to losses from the independent variable by its
standard deviation and then dividing it by the mean value of losses for the period. The advantage
of this calculation is that it allows for comparison across the two periods.
Table 8 reports these marginal effects. These effects are all relatively small. For example,
in the earlier period, a one standard deviation increase of the Interest Receivable Ratio is
associated with an increase of the loss ratio by only 0.65%. The corresponding effect for the later
period is a similarly small 0.53%. In the later period, CLD lending gets up to a still small 1.82%
increase in the loss ratio. In future work, we will check whether measuring the marginal effects
at distributions other than the mean identify nonlinear effects.

7. Counterfactual Exercises
The regression results show that CLD and other real estate lending played an important
role in both banking crises. In this section, we assess whether the large change in bank
characteristics, such as the increase in real estate lending, between the two periods can account
for failure probabilities and the large difference in FDIC losses. We do this by running a
counterfactual exercise where we take the independent variables in each period and calculate the
losses conditional on failure using the estimated coefficients in the other period. 24 This exercise
tells us if changes in bank characteristics and differences in the severity of macroeconomic
shocks can account for differences in failure probabilities and FDIC losses.
Table 9 reports the counterfactual exercises for bank failure rates. For this exercise, we
simply used the probit selection equation results. The first observation is that macroeconomic
effects (the unemployment and house price variables) are important. The actual failure rate for
our sample over 1986-92 was 5.7%. Under the counterfactual where bank characteristics stay the
same but macroeconomic conditions are those from the later period, the failure rate would
increase to 11.6%. Similarly, the actual failure rate in the 2007-13 period was 4.7%. If bank
characteristics in this later period stay the same, but macroeconomic conditions are those from
the earlier period, the failure probability would drop to 1.4%.
The surprising finding is that changes in bank characteristics from 1985 to 2006, which
includes the increased concentration in real estate lending, do not increase failure rates but
instead actually decrease them. This can be seen in the last row of both columns in Table 9. First,
in the 1986-92 period, when macroeconomic conditions are left unchanged but bank
characteristics are changed to those of 2006 Q4, the failure rate drops to 3.3%. Similarly, in the
later period, when only macroeconomic conditions are changed to those of 1985 Q4, the failure
rate drops to a very low 1.4%. Despite the increased concentrations in CLD and other real estate
lending, which do increase failure probabilities, banks in the 2006 Q4 period had other
characteristics that greatly reduced failure probabilities. The most important such characteristic is
capital. Capital is negative and statistically significant for predicting failure probabilities.
Average capital in the later period was around 11%, compared with a smaller 8.5% in the earlier
period. The higher average capital more than offsets the increased risk from the real estate
24

For all counterfactuals where bank characteristics in one period are applied to estimated coefficients in the other
period, we adjust bank size by the ratio of the average size between the two periods. We make this adjustment
because bank size is a nominal variable and average size of banks, as well as total assets in the banking sector, grew
between the two periods. Roughly, the size adjustment for a 1985 Q4 bank operating in the 2007-13 environment is
on the order of three, that is, a $1 billion bank in 1985 Q4 is treated as a $3 billion bank in 2006 Q4 and vice versa.

concentrations. Meanwhile, there are other variables, such as lower C&I concentrations, higher
core deposit ratios, and lower non-performing loan ratios that reduce failure probabilities for
2006 banks relative to 1985 banks.
Capital is particularly valuable in the earlier period. We ran a simple counterfactual
where we uniformly increased each bank’s capital level to get a sense of its role in reducing
failure probabilities. If capital is raised 250 basis points in the earlier period, to get capital close
to the later period’s averages, the failure probabilities would drop to 3.6%. While still valuable in
the later period, a similar increase would only drop failure probabilities to 4.1%. The difference
is due to the significantly larger negative coefficient that was estimated on capital in the earlier
period.
In contrast to failure probabilities, we find little effect from the counterfactuals on loss
ratios. Table 10 shows these results. In the earlier period, actual loss ratios are 20.3% and
changing macro conditions raises it to only 24.1%. Similarly, actual losses in the later period are
28.6% and changing the macroeconomic conditions actually raises it slightly to 30.4%. Here,
changing bank characteristics in the later period to those of 1985 Q4 actually raises losses to
37.1%. One important factor seems to be the constant terms, which is higher in the later period.
The lack of difference in predicted losses from changes in bank characteristics is
evidence that the change in bank commercial real estate concentrations does not explain what is
driving the big change in FDIC losses. Our analysis suggests that if a bank fails, then losses will
be high regardless of the time period, albeit higher in the later period. So by this metric PCA was
ineffective. However, our analysis also found that changes to bank characteristics, particularly
the higher capital levels, reduced failure probabilities, so the degree to which these changes are
attributable to PCA suggests that PCA was effective on this dimension.

8. Conclusion
This paper compared bank failure rates and FDIC losses on failed banks between two
banking crises, 1986-92 and 2007-13. We estimated a two-stage Heckman selection model for
each period to assess the causes of bank failures and the causes of FDIC losses. We found that
for community and mid-sized banks, the two crises were very similar. Virtually identical
variables predicted bank failure as well as FDIC losses, though the size of estimates differed.
Capital, security holdings, CLD lending, non-performing loans, core deposits, house price drops,
and unemployment increases all significantly predicted bank failure in both periods.
20

Interestingly, an accounting variable that is a relatively small asset on the balance sheet, the
Interest Receivable Ratio, predicted both bank failures and higher FDIC losses.
Our analysis finds that the primary difference between the two crises was the size of the
economic shocks, and that these were the biggest factors in determining failure probabilities. The
often noted increased concentration of real estate lending also contributed to increased bank risk
in the recent crisis, but this was offset by the less noted higher capital levels of banks.
In contrast, in our counterfactual experiments, neither changes in bank balance sheets nor
state-level economic shocks explain the large increase in FDIC losses between the earlier and
later period. We also found that, while on average PCA led to banks being shut down before they
had negative book capital, capital levels before failure were only about 300 basis points higher
on average in the later period, so this extra amount of capital was not enough to absorb an
appreciable amount of losses on banks that were put into receivership. It is possible that demand
for bank assets is a factor influencing the value of a failed bank’s assets. Additional data
collection and empirical work would be required to address the demand side for failed bank
assets.

References

Aubuchon, Craig P. and David C.Wheelock. (2010) “The Geographic Distribution and
Characteristics of U.S. Bank Failures, 2007-2010: Do Bank Failures Still Reflect Local
Economic Conditions?” Federal Reserve Bank of St.Louis Review, (September/October),
395-416.
Bennett, Rosalind L. and Haluk Unal. (2014). “Understanding the Components of Bank Failure
Resolution Costs.” Federal Deposit Insurance Corporation Center for Financial Research,
WP 2014-04.
Bennett, Rosalind L. and Haluk Unal. (2014). “The Effects of Resolution Methods and Industry
Stress on the Loss on Assets from Bank Failures.” Journal of Financial Stability.
Forthcoming.
Bovenzi, John F. and Arthur J. Murton. (1988). “Resolution Costs of Bank Failures.” FDIC
Banking Review, vol. 1 (September), 1-13.
Cole, Rebel A. and Jeffrey W. Gunther. (1995). “Separating the Likelihood and Timing of Bank
Failure.” Journal of Banking & Finance, vol. 19 (September), 1073-1089.
Cole, Rebel A. and Lawrence J. White. (2012). “Déjà vu All Over Again: The Causes of U.S.
Commercial Bank Failures This Time Around.” Journal of Financial Services Research,
vol. 42, 5–29.
Fenn, George W. and Rebel A. Cole. (2008). “The Role of Commercial Real Estate Investments
in the Banking Crisis of 1985-1992.” SSRN Working Paper.
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1293473.
Federal Deposit Insurance Corporation. (1998). Managing the Crisis: The FDIC and RTC
Experience 1980-1992. Washington, D.C.
General Accounting Agency. (2013) “Causes and Consequences of Recent Bank Failures.”
Report to Congressional Committees, (January), GAO-13-71.
Heckman, James J. (1979). “Sample Selection Bias as a Specification Error.” Econometrica, vol.
47 (January), 153-161.
James, Christopher. (1991). “The Losses Realized in Bank Failures.” Journal of Finance, vol. 46
(September), 1223-1242.

Jin, Justin Yiqiang, Kiridaran Kanagaretnam, and Gerald J. Lobo. (2011). “Ability of Accounting
and Audit Quality Variables to Predict Bank Failure During the Financial Crisis.”
Journal of Banking & Finance, vol. 35, 2811-2819.
Osterberg, William P. and James B. Thomson. (1995). “Underlying Determinants of ClosedBank Resolution Costs.” In Cottrell, Allin F., Michael S. Lawlor, and John H. Wood
(Eds.) The Causes and Costs of Depository Institution Failures. Kluwer Academic
Publishers: Boston, 75-92.
Whalen, Gary. (1991). “A Proportional Hazards Model of Bank Failure: An Examination of Its
Usefulness as an Early Warning Tool.” Federal Reserve Bank of Cleveland Economic
Review vol. 27 (Quarter 1), 21-31.
White, Lawrence J. (1991) The S&L Debacle (Oxford University Press, New York).

Figure 1 - Commercial bank failures, 1986-2013
Numbers of failed commercial banks in each year from 1986 to 2013.

Bank Failures from 1986-2013
300
250
200
150
100
50
0
1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012
Number of Failures

Figure 2 –Distribution of losses on failed banks, 1986-92 and 2007-13
The graphs below show the distributions of losses in each period for all failed banks except de novo banks.
1986-92

Percent
10

Percent
6

2007-13

.4
Loss_Ratio

Figure 3 – Average capital ratio for all failed banks in the 16 quarters prior to failure
This figure shows the average capital ratio for all failed banks from one to 16 quarters prior to
failure. The capital ratio is defined as total equity capital divided by total assets. The solid lines
represent banks in the later period and the dashed lines represent banks in the early period. Both
de novos and non-de novos are included in this graph as well as banks formed during the sample
periods that failed before the end of it.

percent

Average capital ratio of all failed banks in the 16
quarters prior to failure

12
10
8
6
4
2
0
-2

-4
2007Q1-2013Q4, equally weighted
1986Q1-1992Q4, equally weighted

2007Q1-2013Q4, weighted by assets
1986Q1-1992Q4, weighted by assets

Table 1 – Average adjusted loss ratios, equally weighted and weighted by assets
This table shows the ratio of FDIC losses to assets net of book equity for commercial banks that
failed between 2007-13 and 1986-92, respectively. The sample is divided into three categories:
all banks, established banks (in existence for more than 20 quarters) and de novo banks (in
existence for 20 quarters or less). The first row shows the equally weighted average, whereas the
second row shows the total losses over total assets for all failed banks.

2007-13
All Banks
(403)

Established Only
(306)

De novos Only
(97)

Equally Weighted

29.7

28.6

33.8

Weighted by
Assets

23.8

23.1

35.2

1986-92
All Banks
(998)

Established Only
(714)

De novos Only
(284)

Equally Weighted

21.9

20.3

26.0

Weighted by
Assets

13.8

13.2

24.7

Table 2a – Asset concentration of failed community and mid-sized banks (% of assets)
Securities and various loan categories reported as a percentage of total bank assets for failed
banks in our sample. In 2006, our sample includes all non de novo banks with less than $50
billion in assets, while in 1986 it is one with less than $14.8 billion in assets. 25
1985Q4

2006Q4

Securities

Agricultural loans

Consumer loans

C&I Loans

Constr. Land Develop. Loans

1-4 Family Real Estate

Multi-Family Real Estate

Non-Farm Non-Residential
Real Estate

Table 2b - Asset concentration of all community and mid-sized banks (% of assets)
Securities and various loan categories reported as a percentage of total bank assets for all banks
in our sample. In 2006, our sample includes all non de novo banks with less than $50 billion in
assets, while in 1986 it is one with less than $14.8 billion in assets.

1985Q4

2006Q4

Securities

Agricultural loans

Consumer loans

C&I Loans

Constr. Land Develop. Loans

1-4 Family Real Estate

Multi-Family Real Estate

Non-Farm Non-Residential
Real Estate

* There is some multi-family real estate lending in 1985 Q4, but it rounds to zero.

The asset limit excludes one failed bank in the early period

Table 3 – Variable definitions
Description, creation, and sources for all variables used in the analysis. Most variables are from Call Reports. All
ratio variables are normalized by either total assets or total loans with the exception of the FHLB ratio which is
divided by domestic liabilities and minority interest. De novo banks are defined as those who were in existence for
20 quarters or less.
Variable Name

Description

FDIC Losses

estimated losses to the FDIC
estimated losses to the FDIC divided
by total assets of bank at time of
failure
Assets of a failed bank minus book
equity at latest filed Call Report
before failure
estimated losses to the FDIC divided
by net assets of bank at time of
failure
commercial and industrial loans
divided by total loans and leases net
of unearned income
(domestic) construction and land
development loans divided by total
(domestic) loans and leases net of
unearned income
agricultural loans divided by total
loans and leases net of unearned
income

Loss Ratio
Total Net
Assets at
Failure
Adjusted Loss
Ratio

C&I Loan Ratio

CLD Loan
Ratio
Agricultural
Loan Ratio

Consumer Loan
Ratio
Real Estate
Loan Ratio
1-4 Family Real
Estate Loan
Ratio
Other Real
Estate Loan
Ratio
Multi-Family
Real Estate
Loan Ratio
Non-Farm/
Non-Residential
Loans Ratio
26

Variable Creation

credit card and other consumer loans
divided by total loans and leases net
of unearned income
loans secured by real estate divided
by total loans and leases net of
unearned income
(domestic) RE loans backed by 1-4
family residential properties divided
by total (domestic) loans and leases
net of unearned income
1-4 family loan ratio and CLD loan
ratio subtracted from total real estate
loans
(domestic) RE loans backed by
multi-family residential properties
divided by total (domestic) loans and
leases net of unearned income
(domestic) RE loans backed by
nonfarm nonresidential properties
divided by total (domestic) loans and
leases net of unearned income

FDIC Losses /
Total Net Assets at Failure

Source
FDIC
Historical
Statistics on
Banking
(HSOB) 26

Assets at failure reported by FDIC –
rcfd3210

FDIC HSOB
FDIC HSOB
and Call
Report

FDIC Losses /
Total Net Assets at Failure

FDIC HSOB

rcfd1766 / rcfd2122

Call Report

rcon1415 / rcfd2122

Call Report

rcfd1590 / rcfd2122
(rcfd2008 + rcfd2011) / rcfd2122 in
early period and (rcfdb538 +
rcfdb539 + rcfd2011)/rcfd2122 in
later period

Call Report

rcfd1410/rcfd2122

Call Report

rcon1430 / rcfd2122

Call Report

RE Loan Ratio-CLD Loan Ratio-14 Family RE Ratio

Call Report

rcon1460 / rcfd2122

Call Report

rcon1480/rcfd2122

Call Report

FDIC Historical Statistics on Banking can be found at https://www2.fdic.gov/hsob/SelectRpt.asp?EntryTyp=30

FHLB Advance
Ratio

Loan Loss
Reserves Ratio
Past Due Loans
Non-Accrual
Loans
Nonperforming
Loans Ratio
Interest
Receivable
Ratio
Size
Capital Ratio

Securities Ratio

Core Deposits
Ratio
Earnings

Peak to Trough

Unemployment
Increase
De novo

FHLB advances divided by liabilities
and minority interest (only in later
period)
allowance for loan and lease losses
plus allocated transfer risk reserves
divided by total loans and leases net
of unearned income
loans 90 or more days past due,
divided by assets
Non-accruing loans, divided by
assets
loans 90 or more days past due plus
non-accruing loans, divided by assets
accrued interest receivable divided
by total assets
the natural logarithm of a bank’s
assets
bank equity divided by assets
the book value of all held-to-maturity
securities plus the fair value of all
available-for-sale securities, divided
by assets

core deposits (gathered in a bank's
demographic area) divided by total
liabilities
net income divided by assets
Difference in peak to trough HPI
values when the maximum occurred
prior to the minimum
Increase in the unemployment rate
during the period when the maximum
occurs after the beginning of the
period
1 if less than or equal to 20 quarters
since birth, 0 otherwise

(rcfdf055 + rcfdf056 + rcfdf057 +
rcfdf058 ) / rcfd2948 or (rcfd2651
+ rcfdb565 + rcfdb566)/rcfd2948
prior to 2006Q3

Call Report

rcfd3123 / rcfd2122 in early period
and (rcfd3123 + rcfd3128)/rcfd2122
in later period

Call Report

rcfd1407/rcfd2170

Call Report

rcfd1403/rcfd2170

Call Report

(rcfd1407 + rcfd1403) / rcfd2170
rcfd2164 / rcfd2170 in early period
and rcfdb556/rcfd2170 in later
period

Call Report

ln(rcfd2170)
rcfd3210 / rcfd2170

Call Report
Call Report

rcon0390 / rcfd2170 in early period
and (rcfd1754 + rcfd1773)/rcfd
2170 in later period
(rcon2215 + rcon6810 +
rcon6648)/rcfd2948 in early period
and (rcon2215 + rcon6810 +
rcon0352 + rcon6648)/rcfd2948 in
later period
riad4340 / rcfd2170
(HPI_Max - HPI_Min)/HPI_Min if
maximum occurred before
minimum in the given time period
Unem_Max – Unem_Min
conditional on maximum occurring
after beginning of period

Call Report

Call Report
Call Report
Corelogic HPI
Data via Haver

Haver
Call Report

Table 4– Summary statistics for 1985 Q4 and 2006 Q4
Summary statistics of variables for our sample.
Panel A – Early Period
1986-92
N
Adjusted Loss
713
Ratio
12662
Size
12662
Capital
12662
Securities
12658
CLD Loan
12658
C&I Loans
Core Deposits 12662
Peak to
12662
Trough
12662
Earnings
NonPerforming
12662
Loans
Agricultural
12658
Loans
1-4 Residential
12658
Properties
Consumer
12658
Loans
Other Real
12658
Estate
12658
LLR
Interest
12662
Receivable
Unemployment
12662
Increase

Mean

Min

P25

0.203
10.690
0.085
0.290
0.027
0.223
0.784

0.124
1.195
0.030
0.145
0.050
0.134
0.105

0
6.923
-0.070
0
0
0
0

0.112
9.891
0.067
0.181
0
0.124
0.733

0.054
0.007

0.095
0.014

0.017

P75

Max

0.193
10.577
0.079
0.276
0.007
0.199
0.801

0.277
11.295
0.095
0.384
0.030
0.298
0.855

0.573
16.506
0.837
0.862
0.606
0.958
1

0
0
-0.201 0.005

0
0.010

0.070
0.013

0.451
0.182

0.020

0.004

0.010

0.021

0.305

0.139

0.193

0.001

0.042

0.212

0.999

0.209

0.137

0.103

0.187

0.292

0.983

0.237

0.139

0.137

0.212

0.311

1.058

0.135
0.015

0.090
0.010

0
0

0.067
0.009

0.122
0.012

0.188
0.016

0.738
0.274

0.009

0.006

0.005

0.008

0.012

0.056

1.185

1.459

0.167

2.333

Median

Table 4 (cont.)
Panel B – Later Period
2007-13
N
Adjusted Loss
306
Ratio
6618
Size
6618
Capital
6618
Securities
6590
CLD Loan
6590
C&I Loans
Core Deposits 6617
6618
Peak to
Trough
6618
Earnings
Non6618
Performing
Loans
Agricultural
6590
Loans
1-4 Residential
6590
Properties
Consumer
6590
Loans
Other Real
6590
Estate
6590
LLR
Interest
6590
Receivable
Unemployment 6618
Increase
6617
FHLB Ratio

Mean

Min

P25

0.285
11.850
0.110
0.220
0.103
0.151
0.748
0.201

0.120
1.290
0.058
0.146
0.118
0.098
0.135
0.119

0.045
7.102
-0.001
0
0
0
0
0

0.201
10.987
0.084
0.115
0.017
0.083
0.683
0.121

0.012

0.015

-0.356

0.006

0.008

0.082

P75

Max

0.281
11.729
0.097
0.194
0.061
0.132
0.766
0.182

0.361
12.580
0.118
0.302
0.150
0.196
0.838
0.284

0.851
17.723
1
0.998
0.905
0.920
0.999
0.587

0.008

0.011

0.015

0.595

0.001

0.003

0.008

0.118

0.131

0.0001

0.017

0.111

0.810

0.251

0.157

0.134

0.228

0.342

0.087

0.098

0.030

0.061

0.111

0.309
0.014

0.139
0.012

0
0

0.217
0.010

0.303
0.012

0.387
0.015

1
0.571

0.008
4.826

0.004
1.584

0
1.267

0.005
3.933

0.007
4.7

0.010
5.967

0.056
9.633

0.041

0.058

0.016

0.064

0.976

Median

Table 5 – Heckman selection model for 1986-92 and 2007-13 periods
Heckman selection model for failure probability and loss ratio in both periods. The sample excludes de novos
(defined as banks in existence for 20 quarters or less) and banks over our asset thresholds. Bank specific financial
ratios are measured in 1985 Q4 and 2006 Q4 respectively. All variables are as defined in Table 4. Standard errors
appear in brackets. Levels of significance are indicated by *, **, and *** for 10%, 5%, and 1%, respectively. The
loss equation represents the estimated loss upon failure, whereas the selection equation is a probit equation that
predicts the probability that a bank will fail.
(1)
(2)
1986-92
2007-13
Loss Equation
Loss Ratio
Size
Securities Ratio
CLD Loan Ratio
C&I Loan Ratio
Core Deposit Ratio
Interest Receivable

Selection Equation

Loss Equation

Selection Equation

-0.02***
[0.004]
-0.04
[0.06]
-0.01
[0.06]
0.04
[0.03]
-0.17***
[0.04]
2.02***
[0.71]

-0.20***
[0.03]
-1.61***
[0.25]
2.55***
[0.48]
1.22***
[0.38]
-2.14***
[0.23]
50.14***
[4.48]
-12.95***
[1.36]
2.88***
[0.22]
-3.33*
[1.74]
7.64***
[1.09]
0.08
[0.41]
-0.80**
[0.42]
0.85**
[0.42]
0.30
[0.37]
-1.60
[2.57]
0.07***
[0.02]
2.00***
[0.56]
-0.05***
[0.01]
11,945
713

-0.04***
[0.01]
-0.14**
[0.07]
0.10
[0.06]
0.02
[0.06]
-0.07
[0.05]
5.30***
[1.57]

-0.03
[0.04]
-1.22***
[0.33]
5.59***
[1.27]
2.91**
[1.32]
-1.62***
[0.26]
40.91***
[14.55]
-2.86**
[1.18]
0.90**
[0.46]
-12.58***
[3.13]
15.94***
[3.38]
1.24
[1.47]
1.86
[1.29]
2.76**
[1.27]
0.76
[1.47]
-10.29
[7.19]
0.13***
[0.04]
-3.57**
[1.39]
-0.05**
[0.02]
6,284
306

Capital Ratio
Peak to Trough
Earnings Ratio
Non-Performing Loans
Agricultural Loans
1-4 Residential
Properties
Other Real Estate
Consumer Loan
Loan Loss Reserve
Unemployment
Increase
Constant
Lambda
Censored Observations
Uncensored
Observations
Wald Statistic
Pseudo R2

0.54***
[0.06]

34.70
0.137

0.80***
[0.09]

73.91
0.204

Table 6 – Heckman loss equation and OLS loss equation for both periods
The loss equation in the Heckman model is compared to an OLS regression that uses the
variables in the Heckman loss equation. Levels of significance are indicated by *, **, and ***
for 10%, 5%, and 1%, respectively.
(1)
1986-92

(2)
2007-13

Loss Equation
Heckman
Loss Ratio
Size
Securities Ratio
CLD Loan Ratio
C&I Loan Ratio
Core Deposit
Ratio
Interest
Receivable
Constant
Adjusted RSquared

Loss Equation
OLS

Loss Equation
Heckman

Loss Equation
OLS

-0.02***
[0.004]
-0.04
[0.06]
-0.01
[0.06]
0.04
[0.03]
-0.17***
[0.04]
2.02***
[0.71]

-0.02***
[0.004]
-0.15***
[0.05]
0.06
[0.06]
0.08**
[0.03]
-0.25***
[0.04]
3.98***
[0.60]

-0.04***
[0.01]
-0.14**
[0.07]
0.10
[0.06]
0.02
[0.06]
-0.07
[0.05]
5.30***
[1.57]

-0.04***
[0.01]
-0.19***
[0.06]
0.20***
[0.04]
0.03
[0.07]
-0.13***
[0.04]
5.34***
[1.57]

0.54***
[0.06]

0.55***
[0.07]
0.15

0.80***
[0.09]

0.74***
[0.09]
0.23

Table 7 – Marginal effects on bank failure rates
Marginal effects produced by the probit selection equation. The failure rate is evaluated by
setting all independent variables at their means. The marginal effect for each variable is then
calculated by increasing it by its standard deviation while keeping the other variables at their
means. Italicized variables are in both the selection and loss equations. Non-italicized are only in
the selection equation. Marginal effects for variables that are significant in the selection equation
are in bold.

1986-92
Failure Rate
Evaluated at Mean

2007-13

1.25%

1.13%
Marginal Effects

Size
-0.60%

-0.11%

-0.58%

-0.43%

0.48%

4.12%

0.64%

1.30%

-0.57%

-0.51%

1.17%

0.64%

1.03%

0.39%

-0.82%

-0.41%

-0.14%

-0.45%

0.59%

0.44%

0.05%

0.58%

-0.31%

1.21%

0.14%

0.24%

0.27%

1.76%

-0.05%

-0.32%

0.37%

0.77%

Securities
CLD Loan
C&I Loans
Core Deposits
Interest Receivable
Peak to Trough
Capital
Earnings
Non-Performing
Loans
Agricultural Loans
1-4 Residential
Properties
Consumer Loans
Other Real Estate
LLR
Unemployment
Increase

Table 8 – Marginal effects on FDIC losses
Effects calculated by multiplying the calculated marginal effect by the standard deviation of its
corresponding independent variable. After this step, the resulting product is scaled by the mean
value of losses in the period considered. This method allows for a side-by-side comparison of the
effects of each variable.

Size
Securities
CLD Loans
C&I Loans
Core Deposits
Interest
Receivable
Peak to
Trough
Capital
Earnings
NonPerforming
Loans
Agricultural
Loans
1-4 Residential
Properties
Consumer
Loans
Other Real
Estate
LLR
Unemployment
Increase

1986-92
Marginal
Effects
-0.59
-0.48
0.13
0.57
-0.54

2007-13
Marginal
Effects
-0.29
-0.56
1.82
0.78
-0.63

0.65
0.53
-0.75
-0.09

0.29
-0.44
-0.51

0.29

0.34

0.03

0.44

-0.21

0.78

0.08

0.20

0.15
-0.03

1.03
-0.33

0.19

0.53

0.54

Table 9 – Counterfactual failure rate exercise results
This table displays actual, predicted, and counterfactual rates of failure for both periods using our
sample. The first row shows the actual failure rate in our sample. The second row shows the
failure rate predicted from the selection equation. The third row displays the failure rate when the
variable coefficients from one period are applied to the bank characteristics in the other. The
fourth row displays the predicted failure rate in both periods with the bank characteristics of
1985 Q4 and the macroeconomic values of the later period. The fifth row displays the failure rate
in both periods with bank characteristics of 2006 Q4 and the macroeconomic values of the earlier
period. In all the counterfactuals where bank characteristics from one period are used with
coefficient estimates from the other period, bank size is adjusted by the ratio of average size in
the two periods.

Probability of Failure
1986-92
5.7
(actual)
5.6
(𝛽𝛽85, 𝑥𝑥85 )
10.8 (𝛽𝛽85, 𝑥𝑥06 )
11.6 (𝛽𝛽85, 𝑥𝑥 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏,85, 𝑥𝑥 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚,06 )
3.3
(𝛽𝛽85, 𝑥𝑥 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏,06 , 𝑥𝑥 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚,85 )

2007-13
4.7 (actual)
4.7 (𝛽𝛽06, 𝑥𝑥06 )
1.3 (𝛽𝛽06, 𝑥𝑥85 )
2.6 (𝛽𝛽06, 𝑥𝑥 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏,85 , 𝑥𝑥 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚,06 )
1.4 (𝛽𝛽06, 𝑥𝑥 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏,06 , 𝑥𝑥 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚,85 )

Table 10 - Counterfactual loss ratio exercise results
This table displays actual, predicted, and counterfactual loss ratios for both periods. The first row
shows the average adjusted loss ratio for our sample. The second row represents the loss ratio
that is predicted using the Heckman results of both dependent variables in both periods. The third
row displays the predicted loss ratio when the variable coefficients from one period are applied
to the bank characteristics in the other. The fourth row displays the predicted loss ratio in both
periods with the bank characteristics of 1985 Q4 and the macroeconomic values of the later
period. The fifth row displays the predicted loss ratio in both periods with bank characteristics of
2006 Q4 and the macroeconomic values of the earlier period. In all the counterfactuals where
bank characteristics from one period are used with coefficient estimates from the other period,
bank size is adjusted by the ratio of average size in the two periods.

Adjusted Loss Ratio
1986-92
20.3 (actual)
23.9 (𝛽𝛽85, 𝑥𝑥85 )
21.0 (𝛽𝛽85, 𝑥𝑥06 )
24.1 (𝛽𝛽85, 𝑥𝑥 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏,85 , 𝑥𝑥 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚,06 )
19.2 (𝛽𝛽85, 𝑥𝑥 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏,06 , 𝑥𝑥 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚,85 )

2007-13
28.6 (actual)
31.3 (𝛽𝛽06, 𝑥𝑥06 )
36.6 (𝛽𝛽06, 𝑥𝑥85 )
37.1 (𝛽𝛽06, 𝑥𝑥 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏,85 , 𝑥𝑥 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚,06 )
30.4 (𝛽𝛽06, 𝑥𝑥 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏,06 , 𝑥𝑥 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚,85 )

Full text of Working Papers (Federal Reserve Bank of Richmond) : Did the Financial Reforms of the Early 1990s Fail? A Comparison of Bank Failures and FDIC Losses in the 1986-92 and 2007-13 Periods, Working Paper 15-05

FRASER