Staff StudiesStaff Studies
Federal Deposit Insurance Corporation
Staff Studies
Report No. 2020-04
Measurement of Small Business
Lending Using Call Reports:
Further Insights From the
Small Business Lending Survey
July 2020
www.fdic.gov/cfr • @FDICgov • #FDICCFR • #FDICResearch
1
Measurement of Small Business Lending Using Call Reports:
Further Insights From the Small Business Lending Survey
Jacob Goldston and Yan Y. Lee
1
Abstract: Aggregate U.S. bank lending to small businesses is often approximated with regulatory data based on loan size
using commercial and industrial (C&I) loans of $1million or less as a proxy. We assess the validity of this proxy in measuring
small business lending using data from the Federal Deposit Insurance Corporation’s (FDIC’s) Small Business Lending Survey
(SBLS), a nationally representative survey of banks. We find that more than 30percent of C&I lending by banks with $1billion
to $10billion in assets is in small loans (less than $1million) to large firms or larger loans to small firms, which contradicts
the core assumption of high correlation between loan size and firm size that underpins usage of the proxy. When we relax
the proxy’s assumption for loan size using parameters generated from the survey data, we find that aggregate small business
lending by banks likely recovered from the 2007–2009 recession (Great Recession) two to four years earlier than indicated
by the proxy. Using $10million in gross annual revenue as a maximum small business size, we find that the proxy on net
understates small business lending by banks with $1billion to $10billion in assets by up to 23percentage points, with
wide dispersion among banks in the degree of over- and understatement. This dispersion undermines the proxy’s ability to
correctly rank banks by the proportion of their C&I loan portfolio devoted to small business lending. We also find that the
proxy likely only mildly understates small business lending for the smallest banks, those with less than $1billion in assets.
1
Economists, Federal Deposit Insurance Corporation.
The views expressed are those of the authors and do not necessarily reflect the official positions of the Federal Deposit Insurance Corporation or
the United States. FDIC Staff Studies can be cited without additional permission.
2
1 Introduction
Small businesses are important to the U.S. economy, and banks are important to small businesses. In 2015, there were
30.2million small businesses in the United States.
2
These firms make up 99.9percent of all U.S. businesses and generate
47.5percent of private sector employment and 40.8percent of private sector payroll. Small businesses typically cite
bank credit as their most common source of external financing.
3
Bank lending to small businesses is monitored through
Consolidated Reports of Condition and Income (Call Reports).
4
Since 1993, C&I loans of $1million or less have been used as
a proxy for small business lending. While the proxy reliably measures small loans to businesses, it does not directly measure
loans to businesses that are small.
5
Using proxy data, the standard narrative of small business lending since the Great Recession is one of a slow recovery. In
a report to the Small Business Administration (SBA), Cole (2018) states that bank lending to small businesses “plummeted by
almost 20percent during 2009–2011” and that “bank lending to small businesses continued to decline during the post-crisis
years of 2012–2015.” Reichow (2017) describes small business lending as “languishing” and notes that “small-business loans
at U.S. commercial banks have declined 15percent from their precrisis high.” Cortes et al (2020) motivate their research on
the effect of stress testing by noting that “lending to large businesses has bounced back since the Great Recession … [t]he
recovery of small business lending, however, continues to be slow.” Bordo and Duca (2018) situate their work on the impact
of the Dodd-Frank Act in a similar way.
These conclusions assume that the proxy is an accurate measure of small business lending. The basic assumptions
required to use the proxy as a measure of small business lending are that small loans (less than $1million) tend to be made
to small businesses and larger loans to large businesses. But had the proxy threshold of $1million been adjusted for inflation,
it would have been more than $1.6million by December 2015.
6
This suggests that loans greater than the $1million threshold
may be extended to small businesses and that the proxy understates small business lending. Conversely, some small loans
may be extended to large businesses, which would result in an overstatement of small business lending when relying on the
proxy. The goal of this study is to determine whether the underlying assumptions of the proxy hold and to investigate the
implications if these assumptions are violated.
Given the proxy’s potential to misstate small business lending, the Government Accountability Office (GAO)
recommended in 2018 that the three federal banking regulatory agencies— the FDIC, the Board of Governors of the Federal
Reserve System, and the Office of the Comptroller of the Currency—“reevaluate, and modify as needed, the requirements for
the data banks report in the Consolidated Reports of Condition and Income to better reflect lending to small businesses.”
7
In October 2019, the agencies published a joint Federal Register notice requesting public comment on ways to modify data
collection to “better reflect lending to these sectors of the U.S. economy.”
8
The implementation of any changes to Call Report
collection is of particular interest to the FDIC given its role as steward of Call Report data.
9
2
“Frequently Asked Questions About Small Business,” U.S. Small Business Administration Oice of Advocacy, U.S. Small Business Administration,
August 2018, https://www.sba.gov/sites/default/files/advocacy/Frequently-Asked-Questions-Small-Business-2018.pdf.
3
Federal Reserve Small Business Credit Survey, Report on Employer Firms, 2017.
4
While bank-level outstanding loan volumes are collected under Call Reports, banks also annually provide data under the Community Reinvestment Act on
the geographic location of their new C&I loan originations of $1million or less.
5
In addition, loans collateralized by residential real estate are not captured by the proxy.
6
Adjustment using the Bureau of Labor Statistics’ Consumer Price Index for All Urban Consumers (CPI-U).
7
“Community Banks: Eect of Regulations on Small Business Lending and Institutions Appears Modest, but Lending Data Could Be Improved,”
U.S. Government Accountability Oice, September 5, 2018, https://www.gao.gov/reports/GAO-18-312/.
8
See 84 Fed. Reg. 55687 (October 17, 2019).
9
Section 122 of the FDIC Improvement Act of 1991 provides that “the appropriate Federal banking agency shall prescribe regulations requiring insured
depository institutions to annually submit information on small businesses and small farm lending in their reports of condition.” In addition, Section 7(a)
of the Federal Deposit Insurance Act provides that Call Reports for state nonmember banks be in the form and contain information that the FDIC Board of
Directors requires.
3
In this study, we use data from the FDIC’s Small Business Lending Survey (SBLS) to evaluate the usefulness of the proxy
in measuring small business lending. The SBLS is a nationally representative survey of U.S. banks’ small business lending
activities and practices. Critical to our study, a subset of the survey questions asked banks with greater than $1billion in
assets to break down their outstanding C&I loan balances by both loan size and borrower firm size, providing more detail than
Call Reports. We use these responses, along with other quantitative and qualitative survey questions, to evaluate the accuracy
and precision of the proxy. Because the SBLS is nationally representative and has a high response rate, it can be used to make
statistical inferences about the entire population of U.S. banks.
Questions regarding banks’ calendar year 2015 activities were asked during data collection in 2016 and 2017, and a report
of main findings was published in 2018.
10
The survey was conducted and implemented by the U.S. Census Bureau (Census)
to ensure the confidentiality of data collected from respondent banks. From a universe of 6,018 institutions, 1,961banks
were sampled
11
and asked to complete the survey and 1,174 ultimately responded, yielding an overall response rate of
59.9percent.
12
The number of large banks (those with assets of greater than $10billion) that responded to the quantitative questions
about loan and firm size fell below the reporting threshold for protecting respondents’ identities, which limits the scope of
this study.
13
Therefore, the main analysis and conclusions primarily use data from banks with $1billion to less than $10billion
(“$1billion to $10billion”) in assets. While no separate analysis is conducted for large banks, responses for these banks are
incorporated if the additional data improve the precision of estimates. We also analyze the accuracy of the proxy for the
smallest banks, those with less than $1billion in assets, using a qualitative survey question on whether banks focus their
C&Ilending activity on small businesses.
Using the survey data combined with some reasonable assumptions, we find that the proxy mildly understates small
business lending by the smallest banks. However, for banks with $1billion to $10billion in assets, the proxy’s underlying
assumption—that small loans are much more likely than large loans to go to small businesses—is contradicted by banks’
actual lending practices. Comparisons between Call Report data and the survey data show that loan size and firm size are
only weakly correlated, which causes the proxy to be both generally inaccurate at the industry level and imprecise at the
bank level. At the industry level, more than 30percent of the C&I lending by banks with $1billion to $10billion in assets is in
small loans to large businesses or large loans to small businesses when using either of the two small business size thresholds
available in the survey data. On net, the proxy likely understates industry-level small business lending for banks with assets of
$1billion to $10billion by up to 23percentage points. Further, at the bank level, we find wide dispersion across banks in how
much the proxy over- or understates small business lending.
10
See FDIC (2018) for more detail on the SBLS.
11
The SBLS sample included banks of all sizes and in both urban and rural areas. Drawing from institutions regulated by all three banking agencies yielded a
universe of 6,018 banks that were FDIC-insured, had filed both a Q4 2015 Call Report and annual Summary of Deposits branch-level geographic data, had at
least one branch in a U.S. state or in the District of Columbia, had positive outstanding loan activity, received core deposits, and were of an entity type likely
to conduct nonzero commercial lending activity. The few institutions that were dropped include bridge banks, industrial banks, investment banks, joint stock
companies, life insurance companies, and pension funds. Within the universe, banks with assets of less than $10billion were stratified along two dimensions
into six strata: (1) total asset size (less than $250million, $250million to less than $1billion, and $1billion to less than $10billion); and (2) whether the
majority of the institution’s deposits were in branches located within a metropolitan statistical area (MSA). Banks with assets greater than $10billion were
stratified, by asset size only, into two additional strata: $10billion to less than $50billion, and $50billion or more. Overall, eight strata were created. See
Appendix A of FDIC (2018).
12
Overall survey response rates were 60.0percent for small banks (those with assets of less than $10billion) and 57.8percent for large banks (those with
assets of $10billion or greater). Across the eight strata, response rates ranged from 56.3percent (banks with $50billion or more in assets) to 63.9percent
(banks in primarily non-MSA areas with assets of $1billion to less than $10billion). See Appendix A of FDIC (2018).
13
The lack of usable data from large banks is unsurprising. Relatively few (90) large banks were in our universe at the time of the survey. Those that decided
to participate needed to have data systems required to answer the questions and to answer each question. While for non-quantitative questions (31 distinct
questions relevant to large banks) the average question-by-question response rates was 93.0percent for banks with $10billion to less than $50billion in
assets and 94.4percent for banks with $50billion or more in assets, large banks exhibited the largest drop-o in average response rates for quantitative
outstanding loan balance questions (16 distinct items). See Appendix A of FDIC (2018).
4
These discrepancies are not inconsequential. Adjusting proxy shares to allow for even a small proportion of large loans
to be made to small businesses significantly alters our understanding of the evolution of small business lending since the
Great Recession. Using survey-generated metrics of banks’ actual lending activity shows that small business lending after
the recession likely recovered two to four years earlier than the proxy indicates. And given the wide variation in the degree to
which the proxy over- or understates bank-level small business lending, the ability of the proxy to correctly identify whether
one bank has a higher share of its commercial loan portfolio in lending to small businesses is limited.
The study is organized as follows: Section 2 discusses how the proxy measure came about and how SBLS data can be
used to assess its potential limitations. Section 3 contains our main analysis of the accuracy and precision of the proxy,
primarily for banks with $1billion to $10billion in assets. We exploit a disaggregation of banks’ C&I lending by firm gross
annual revenue (GAR) not collected by Call Reports but available in the survey data. Section 4 presents a separate analysis of
the overall accuracy of the proxy for measuring small business lending for only the smallest banks, those with assets of less
than $1billion. Section 5 concludes.
2 The Proxy and Diiculties in Measuring Small Business Lending
Understanding bank lending to small businesses is critical given the importance of small businesses to the U.S. real
economy and the importance of banks to small businesses. Unlike large businesses, small businesses generally do not have
access to capital markets and, therefore, typically rely on banks for most of their external financing.
14
Also, it is believed that
smaller banks are more suited to meet the credit needs of smaller businesses, often considered opaque given their lack of
audited financial statements.
15
Therefore, Congress mandated that small businesses’ dependence on banks be monitored,
which regulators implemented through the collection of proxy data via Call Reports.
16
To determine whether the proxy is a
useful measure of small business lending, we first discuss how the current proxy came about, introduce the SBLS survey
questions that can be used to investigate the proxy’s potential limitations, and evaluate whether survey data on firm size
reflect banks’ actual lending to small businesses.
2.1 A History of the Proxy for Small Business Lending
U.S. banks must file Call Reports at the end of each quarter. Section 122 of the FDIC Improvement Act of 1991 required
the banking regulatory agencies to annually collect data on bank lending to small businesses.
17
This requirement was
implemented in 1993 by the new Schedule RC-C, Part II of the Call Report.
18
Banks must provide “outstanding loan balances
with original amounts of $1million or less for commercial and industrial purposes, excluding loans secured by real estate and
loans for the purpose of financial agricultural production.” Banks provide the total number and total amount outstanding
for these loans, stratified into three buckets based on loan amount at the time of origination: less than $100,000, $100,000 to
14
For example, the Kauman Firm Survey, an annual longitudinal study that has followed almost 5,000 new firms since 2004, finds that outside debt, such
as credit cards, credit lines, and bank loans, was the most important source of financing for new firms, followed by owner equity (Robb and Robinson, 2010).
Combined, these two sources account for about 75percent of startup capital.
15
See Udell (2008) for a discussion of nonfinancial statement-based relationship lending technologies. Small banks are believed to be able to more cost
eectively process non-quantifiable so information, perhaps given their closer physical proximity to borrowers or flatter organizational structures, to make
viable loan decisions for borrowers that may have diiculty receiving financing elsewhere. See, for example Berger et al. (2005) and Berger and Udell (2002).
16
See 84 Fed. Reg. 55688-9 (October 17, 2019) for more background on how the Federal Reserve Board uses these data to monitor credit conditions and
assess real economic activity.
17
Banks must collect data on small farm lending as well; see 12 U.S.C. § 1817.
18
See 57 Fed. Reg. 54633 (November 19, 1992). The schedule was to be completed as of June 30 of each year. Aer the Great Recession, to collect better trend
data, the agencies increased the reporting frequency to quarterly beginning with the first quarter 2010 Call Report; see 74 Fed. Reg. 68322 (December 23,
2009). In 2017, the agencies again adjusted the frequency of the collection for institutions with less than $5billion in assets, from quarterly to semiannually,
to reduce reporting burden. Banks with assets of less than $5billion submit the “Consolidated Reports of Condition and Income for a Bank with Domestic
Oices Only and Total Assets Less than $5 Billion,” or FFIEC 051. The two other versions of the Call Reports are the “Consolidated Reports of Condition and
Income for a Bank with Domestic and Foreign Oices,” or FFIEC 031, and the “Consolidated Reports of Condition and Income for a Bank with Domestic Oices
Only,” or FFIEC 041. See 82 Fed. Reg. 2444 (January 9, 2017).
5
$250,000, and $250,000 to $1million. Throughout this study, we will refer to the dollar volume of this collection as the proxy
for small business lending.
19
The proxy came about because of perceived difficulties banks had in providing data by business size. In 1992, the Federal
Financial Institutions Examination Council (FFIEC) requested comment on proposed reporting requirements that based the
collection of small business loan data on a business’s annual sales; the FFIEC proposed that the “appropriate upper limit in
defining the size of a small business” be $10million in annual sales.
20
These loans were to be separately reported for three size
categories of small businesses: those with less than $250,000 in annual sales, those with $250,000 to $1million in annual sales,
and those with $1million to $10million in annual sales. Consistent with existing reporting requirements, the FFIEC proposed
that banks report the number and amount outstanding of C&I loans or loans secured by nonfarm nonresidential properties.
The industry response to the proposal was negative: 436 of the 537 comments received from banks, thrifts, and their
trade associations opposed the proposed reporting requirements. Depository institutions indicated that while annual
sales may be included in a borrower’s credit file, sales data “are not usually contained in either automated or manual loan
systems.”
21
Bank loan systems would therefore have to be modified to meet the proposed requirement. Some commenters
suggested using a measure based instead on loan size, which was already available in their loan systems and which they
believed would be indicative of borrower size.
In response to the public comments, the FFIEC adopted a loan-size approach to measure small business lending, since
this approach would “allow institutions to report information of comparable value at a lower cost to the industry.”
22
The FFIEC
also said the 1989 National Survey of Small Business Finance confirmed a “strong correlation between size of business and
loan size.”
The FFIEC deemed a loan-size cutoff of $1million as “reasonable.” At the time, there were more than 9,500 depository
institutions with less than $100million in assets; $1million was about the maximum amount these banks could lend to a
single borrower given the combination of legal lending constraints and bank equity capital positions.
23
Further, $1million
was the amount suggested by commenters who had advocated using a measure based on loan size. Therefore, federal bank
regulators adopted and still use outstanding balances for loans of up to $1million at the time of origination as a proxy for
small businesslending.
24
The SBLS allows us to explore banks’ use of technology to determine if institutions today are better equipped to report
loans by firm size and if bank size affects an institution’s ability to report that data. Banks were asked whether their core data
systems could collect information related to loan volume by firm size. The findings show that banks still do not use data
systems that collect and can report lending by firm sales volume. The ability to provide outstanding loan balances—data
reported in the Call Report—by firm gross annual revenue is lowest for the smallest banks and increases with bank size
(Figure1). Only 12.3percent of banks with less than $250million in assets can report loans for firms with GAR of $1million
or less and only one-tenth can do so by any firm size. Even among large banks, only 47.9percent can report loans for firms
19
In addition to small commercial and industrial loans, banks report small agricultural loans and small loans secured by nonfarm nonresidential properties.
In some instances, such as 84 Fed. Reg. 55688 (October 17, 2019), the definition of “loans to small businesses” includes these other categories as well.
Throughout the paper, we refer solely to small commercial and industrial loans for consistency between Call Report data and SBLS data.
20
See 57 Fed. Reg. 21409–21414 (May 20, 1992) for the originally proposed reporting requirements presented for public comment. Comment was also
requested on whether reporting institutions should be allowed to use a three-year average for annual sales and whether the size category of the small
business be fixed to the time of the credit decision.
21
See 57 Fed. Reg. 54235–54239 (November 17, 1992) for the FFIEC’s summary of public comments received and the final approved reporting requirements.
22
See 57 Fed. Reg. 54237 (November 17, 1992).
23
Legal lending limits typically allow banks to make loans to one borrower up to 15percent of equity capital (more if secured by readily marketable
collateral). Therefore, a $100million bank with a 10percent equity capital ratio could lend up to $1.5million to one borrower. At the end of second quarter
1992, the ratio of equity capital to total assets for commercial banks with $100million in assets was 9.4percent, implying that the smaller amount of
$1million was reasonable for these institutions. See 57 Fed. Reg. 54238 (November 17, 1992) for more discussion on the FFIEC’s reasoning.
24
The $1million loan-size cuto is likely outdated, however. By fourth quarter 2015, the average legal lending limit for the almost 6,000 banks with less than
$10billion in assets was $8.9million. Calculations were made using Call Report data and reference only the portion of the limit for unsecured lending.
6
witha GAR of $1million or less and 27.0percent can report loans by any firm size. Within each bank size group, banks are only
slightly more likely to be able to provide data by firm GAR for new loan originations relative to outstanding loan balances.
25
These findings suggest that the vast majority of banks likely would face transition costs if they had to collect data on bank
loans to small businesses based on business size. But using loan size as a proxy for small business lending also has a cost in
terms of the current proxy’s potential to inaccurately and imprecisely measure small business lending.
Figure 1
My Bank’s Core Data Systems Have the Ability to Retrieve the
Number and Dollar Volume of…
(Percent Bank-Weighted)
32.8
53.7
27
47.9
28.5
46.5
22.5
40.9
20.4
33.2
19.7
30.4
12.7
14.4
10
12.3
020406080 100
...new loan originations and
renewals, by firm gross annual
revenue, for any sized firm?
(Screener 3)
...new loan originations and
renewals, made to firms with
$1 million or less in gross
annual revenue? (Screener 2)
...outstanding loan balances,
by firm gross annual revenue,
for any sized firm? (Screener 7)
...outstanding loan balances,
made to firms with $1 million or
less in gross annual revenue?
(Screener 6)
$10B+$1B to Less Than $10B$250M to Less Than $1BLess Than $250M
For Calendar Year 2015
As of Call Report Date
December 31, 2015
Source: Small Business Lending Survey (SBLS) Screener Questions S2, S3, S6, and S7.
Note: See Appendix C of the 2018 SBLS report for the survey instrument. For each screener question, dierences
across bank size groups are generally statistically significant at the 10 percent significance level, with most
dierences significant at the 1 percent significance level. Within each bank size group, dierences between
outstanding loan balances and new originations for the same firm size are generally statistically significant at
the 10 percent level. SBLS responses are bank-weighted to represent the proportion of banks in the U.S. banking
industry that exhibit these characteristics.
2.2 Business Size Versus Loan Size
Because the proxy is based entirely on loan size, there are two ways in which it could deviate from true small business
lending. First, banks may make small loans to large businesses; these loans would be included in the proxy but are not small
business lending. Second, banks may make large loans to small businesses; these loans would be excluded from the proxy
even though they are small business lending. Figure2 depicts the lending to small businesses that regulators hope to capture
(left column) compared with the actual lending captured by proxy available in Call Reports (top row). Small businesses are
those firms below a particular size threshold (typically gross annual revenue, in the context of bank lending) or based on
certain criteria. These firms receive loans of varying sizes, some above and some below the $1million proxy threshold. The
proportion of total business lending by a bank (or by the banking industry) that is true “small business lending” is represented
25
Within each small bank size group, the dierences in banks’ ability to provide data by firm size between outstanding loan balances and new originations are
typically statistically significant at no less than the 10percent significance level; the dierences within the large bank group are marginally significant. Banks
may be better able to provide originations data by whether the firm has gross annual revenue of $1million or less because of Community Reinvestment Act
(CRA) requirements for banks of above about $1billion in assets (asset thresholds for reporting are CPI-adjusted yearly). To verify whether banks lend back
into the communities from which deposits are received, these banks are asked for the geographic distribution of new C&I origination dollars for loans of
$1million or less. While banks are not required to provide these data by firm GAR, they are asked to provide the volume of the subset that is to firms with GAR
of less than $1million, if they already collect these data of their own accord. See the FFIEC’s “A Guide to CRA Data Collection and Reporting.”
7
by Cell A and Cell B, which together equal the share composed of loans of any size extended to a small business. By contrast,
the proxy measures small loans to businesses; this proportion is represented by the sum of Cell A and Cell C. Therefore, Cell B
is the gross understatement of small business lending when using the proxy (large loans extended to small businesses), and
Cell C is the gross overstatement (small loans extended to large businesses).
Figure 2
Small Business Lending Versus Call Report Proxy Lending
(As Share of C&I Loan Portfolio)
Larger
Businesses
CA
Proxy
Lending
(A+C)
BD
True Small
Business Lending
(A+B)
Small
Businesses
$1 Million
or Less
Greater Than
$1 Million
Source: Authors’ illustration.
Business
Loans of
Loans to
To assess the proxy’s accuracy in capturing lending to small businesses, we need to know the sizes of businesses
receiving the loans. In this study, we use an SBLS survey question that asked banks with $1billion or more in assets to allocate
their year-end 2015 Call Report C&I outstanding loan dollars both for loans less than $1million at the time of origination (the
proxy) and the remainder that was above $1million, by three firm gross annual revenue buckets:
1. $1million or less,
2. Greater than $1million to less than $10million, and
3. $10million or more.
Therefore, two possible small business thresholds are available from the survey data: GAR of $1million or less and GAR
of less than $10million. FDIC (2018) assumes that all firms with GAR of $1million or less are small businesses to show that a
nonzero share of loans not captured by the proxy were extended to those businesses—that the share represented by Cell B
from Figure2 is greater than zero. In dollar terms, at banks with assets of $1billion to $10billion, FDIC (2018) found at least
$19.1billion in gross understatement of small business lending (in which small businesses with less than $1million in gross
annual revenue received loans with amounts greater than $1million).
26
26
Values in 2015 dollars. The estimates in FDIC (2018) dier somewhat from those in this study because of dierences in how banks that gave partial answers
to certain SBLS questions were treated.
8
However, gross overstatement may also exist in the proxy, represented by Cell C in Figure2. While FDIC (2018) addresses
only gross understatement based on the $1million GAR threshold, here we investigate overall net misstatement by
considering both gross overstatement and gross understatement in the proxy. To further our understanding of the potential
limitations of the proxy as a measure of small business lending, we use both small business size thresholds available from the
SBLS data ($1million GAR and $10million GAR) to generate our estimates.
2.3 What Size Is a Small Business, and Does It Matter for Banks?
One of the fundamental difficulties in measuring small business lending is that there is no single, consistent definition of
a “small business” or even a single agreed-upon metric.
27
Reporting requirements under the Community Reinvestment Act
(CRA) suggest a GAR threshold of $1million, while the SBA uses industry-based size standards based partially on GAR and
partially on other metrics. Also, less than one-half of banks define small business borrowers internally using gross annual
revenue. Here, we describe some of the GAR sizes commonly used for regulatory purposes to define small businesses and
present survey results from the SBLS to show the size thresholds most often used by banks—if they use such a threshold.
Even accepting GAR as the proper metric, we show that a threshold of $1million misses a substantial quantity of self-defined
small business lending by banks. Therefore, in the main analysis of this study we use both the $1million and $10million
GARthresholds.
2.3.1 No Consistent Definition for Small Businesses, Even for Regulators
Federal regulators use many size thresholds to define a small business. The FFIEC initially suggested in 1992 that firms
with annual sales of less than $10million per year be considered small businesses for the purposes of bank lending. In the
CRA collection, if a bank already collects firm gross annual revenue data, it is then asked to report the volume of loans made
to firms with GAR of $1million or less—the idea being that data by firm size would “enable…the public to judge whether
the institution is lending to small businesses…or whether it is only making small loans to larger businesses.”
28
However, in
findings and recommendations provided to the FDIC gleaned from bank field testing interviews conducted to develop the
SBLS, the Census noted that “many respondents…disagree with the most-known definition of small business, $1,000,000 in
annual revenue” and that usage of this threshold “seems to be increasingly rare.”
29
Consistent with this sentiment, the SBA uses business size thresholds typically several magnitudes greater than $1million
firm GAR to conduct their small business guarantee programs.
30
The SBA uses 31 industry-based size standards, 16 of which
are receipts-based thresholds.
31
Of those 16 thresholds, only one is set at $1million; all others are set at $6million or above
and range up to $41.5million. For example, $7.5million is the size standard for the services, retail trade, and construction
industries. Since 2009, the SBA also allows an alternative non-industry–based size standard for its 7(a) and 504/Certified
Development Company (CDC) programs: no more than $15million in tangible net worth and no more than $5million in
average net income after taxes.
27
There are many dierent ways to characterize a small business besides business size. More than one-half of banks with less than $10billion in assets
(47.9percent) and a majority of banks with more than $10billion in assets (78.9percent) use revenue or sales to characterize their small business borrowers.
Banks with less than $10billion in assets also oen cite aspects such as ownership structure, number of employees, footprint in the local community, and
owner involvement in the day-to-day operations of the business. See Section II and SBLS Question 1 in Appendix C of FDIC (2018).
28
See page 13 of the FFIEC’s “A Guide to CRA Data Collection and Reporting.” In practice, the CRA data collected by firm size are not informative for many
banks. No separate indicator exists for whether a bank cannot provide data by firm size; therefore, relying on the indicator for loans to businesses with
$1million GAR or less would make banks that cannot report by firm GAR appear from their CRA data to have extended no loans to those businesses.
29
Field testing of the SBLS survey instrument was conducted throughout 2015 over three rounds with 40 banks of all size groups, in both rural and urban
areas of ten states. The testing showed that community banks typically did not use size criteria to classify commercial loans as small business loans, but
rather considered that “most business loans were for small businesses because of the size of the banks” themselves. See Appendix A of FDIC (2018).
30
The SBA guarantees loans made through banks, credit unions, and other lenders to U.S. small businesses that would have diiculty obtaining credit
without its assistance. See 13 U.S.C. § 121.
31
There are also nine employee-based size standards, one asset-based standard, and one size standard based on the combination of number of employees
and “barrel per day refining capacity.” See Dilger (2019).
9
2.3.2 Small Business Sizes Used by Banks
The actual business size thresholds used by banks may differ from those used by government entities. The SBLS allows
us to determine whether banks formally use firm gross annual revenue thresholds to conduct their small business lending,
and if they do, what those specific thresholds are.
32
Figure3A shows that larger banks are more likely to formally use a GAR
threshold to define small businesses. Less than one-fifth (18.7percent) of the smallest banks use such a threshold compared
with more than one-third (35.7percent) of banks with $1billion to $10billion in assets.
33
Further, while a majority of large
banks (61.8percent) use a GAR threshold, about two-fifths do not. For the banks that use a formal GAR threshold, the majority
of banks of all sizes use a threshold above $1million in firm gross annual revenue (Figure3B). Larger banks typically use
higher thresholds. For banks with less than $1billion in assets, 25.1percent of those that use such a threshold use one greater
than $5million in firm GAR, while 37.0percent of banks with $1billion to $10billion in assets that use a threshold use one of
that size.
Figure 3A
Share of Banks That Use GAR Thresholds
(Percent Bank-Weighted)
18.7
35.7
61.8
0
20
40
60
80
100
< $1B $1B to < $10B $10B+
Percent
Source: Small Business Lending Survey (SBLS) Question 3.
Note: GAR is gross annual revenue. See Appendix C of the 2018 SBLS report
for the survey instrument. Dierences between bank size groups are
statistically significant at the 1 percent level. SBLS responses are bank-
weighted to represent the proportion of banks in the U.S. banking industry
that exhibit these characteristics.
Figure 3B
Share of Banks Using Each GAR Threshold
Among Banks That Use GAR Thresholds
(Percent Bank-Weighted)
0
20
40
60
80
100
< $1B $1B to < $10B $10B+
Percent
48.2
34.2
15.3
26.7
28.8
28.2
25.1
37.0
56.4
<= $1M > $1M to <= $5M > $5MGAR Threshold Used:
Source: Small Business Lending Survey (SBLS) Question 3a.
Note: GAR is gross annual revenue. See Appendix C of the 2018 SBLS report for the
survey instrument. For banks with less than $1 billion in assets, there are statistically
significant dierences at the 5 percent significance level between the two larger GAR
thresholds and the GAR threshold of <=$1 million. For banks with $1 billion to $10 billion
in assets, there are no statistically significant dierences in the use of each of the three
GAR thresholds. For banks with $10 billion or more in assets, there are dierences at
the 10 percent significance level between the >=$5 million GAR threshold and the two
smaller GAR thresholds. SBLS responses are bank-weighted to represent the
proportion of banks in the U.S. banking industry that exhibit these characteristics.
2.3.3 Does Small Business Lending Occur Above the $1 Million GAR Threshold?
Using a formal size threshold to identify whether a business is small does not necessarily imply that lending is actually
extended to businesses of that size. We use survey data to confirm that banks extend loans of any size to businesses that they
consider small but are larger than the $1million GAR threshold. The SBLS asked banks about their new origination volumes in
relation to internal bank thresholds. Banks with $1billion or more in assets were asked for the dollar volume of their C&I loans
to businesses with gross annual revenue of $1million or less, regardless of the internal threshold, and C&I loans to businesses
with gross annual revenue of less than the bank’s internal GAR threshold (if the bank had indicated in an earlier question that
a formal threshold was used).
34
32
See SBLS Question 3 and SBLS Question 3A in Appendix C of FDIC (2018).
33
FDIC (2018) shows that only 20.4percent of banks with less than $10billion in assets use firm gross annual revenue thresholds to define small businesses.
34
See SBLS Question 8 in Appendix C of FDIC (2018). The definition of C&I lending diers from the Call Report definition in that it includes all loans for C&I
purposes, regardless of how the loans were secured. Banks were also asked the volume of these C&I new originations primarily secured by 1–4 family
residential properties. FDIC (2018) documents that at least $18billion in outstanding small business C&I lending was not measured by the proxy because the
loans were residentially secured (this lending is captured in the Call Report as home mortgage loans).
10
We find that banks do a considerable amount of lending to firms with GAR above $1million but below their internal
thresholds. Banks with $1 billion to $10billion in assets that use a formal GAR threshold made a total of 29percent of their
C&I loans to businesses that they consider small: 18.7percent to firms with GAR below $1million and 10.3percent to firms
with GAR above $1million but below their internal threshold (Table1). This means that more than one-third of self-described
small business lending by these banks is to firms that would be excluded under the $1million GAR definition. If we broaden
our scope to include banks with more than $10billion in assets, we find that nearly two-thirds of self-described small business
lending made to firms with GAR above $1million.
Table 1
Share of New C&I Originations
Banks With $1 Billion or More in Assets That Use a Formal GAR Threshold
(Percent Loan-Weighted)
Firm GAR
Small Banks All Banks
$1B to Less Than $10B $1B or More
Less Than $1 Million 18.7% 6.6%
Between $1 Million and Internal Bank GAR Threshold 10.3% 13.0%
Source: Small Business Lending Survey (SBLS) Questions 8, 10, and 11.
Note: GAR is gross annual revenue. See Appendix C of the 2018 SBLS report for the survey instrument. SBLS responses are
loan-weighted to represent the proportion of U.S. industry lending for banks of these sizes that use a formal GAR threshold.
Our working assumption from these findings is that for the typical bank, a GAR threshold of $1million is overly
conservative and would exclude many firms that should properly be considered small businesses. A GAR threshold of
$10million (the original small business size metric proposed by the FFIEC in 1992) would safely capture these larger small
businesses but may include larger firms that certain banks might not consider small.
35
Given the lack of consensus on the
correct definition of a small business, we present results using both thresholds wherever possible.
3 Assessment of the Accuracy and Precision of the Proxy in Measuring
Small Business Lending
In this section, we assess whether and to what degree the proxy diverges from its underlying assumptions (that small
loans are made to small businesses and large loans to large businesses). We do this by filling in the cells of Figure2 using
data from the SBLS, apportioning small and large C&I loans by firm size. We then explore the implications of any inaccuracy
if relying on the proxy as a measure of actual small business lending. Finally, we investigate the dispersion of the bank-level
difference between the proxy and true small business lending, and whether the proxy can properly identify which of two
individual banks makes a larger proportion of their C&I loan dollars to small businesses.
Only banks with $1billion or more in assets were asked the survey questions that we rely on in this section.
Unfortunately, the number of large banks—those with assets greater than $10billion—that responded to these questions fell
below the allowable reporting threshold for protecting respondents’ identities. Therefore, our analysis is primarily limited to
banks with $1billion to $10billion in assets.
36
No separate analysis is provided for large banks; however, responses for large
banks are sometimes incorporated if the additional data improve the precision of estimates.
35
Using the $10million GAR threshold from the SBLS data does not imply that we are measuring actual lending to businesses of this size but rather that we
are capturing lending to any business below this size.
36
As of year-end 2018, these banks held 10.5percent of outstanding domestic C&I loans and 15.9percent of outstanding proxy loans.
11
3.1 Accuracy of the Proxy in Measuring Small Business Lending
We find that relying on the proxy to measure small business lending results in considerable gross misstatement
(combining both overstatement and understatement), regardless of which GAR threshold is used to define small businesses.
Table2 shows the proportion of outstanding loan balances within each of the cells of Figure2 for banks with $1billion to
$10billion in assets, using the $1million GAR threshold in panel A and the $10million threshold in panel B.
37
With either
threshold, we find substantial gross misstatement, with more than one-third of outstanding C&I balances in the form of large
loans to small businesses (Cell B, gross understatement) or small loans to large businesses (Cell C, gross overstatement). If the
proxy were a perfect measure of small business lending, then these cells would both equal zero.
In addition to this gross misstatement of small businesses lending, we also find substantial net misstatement, with the
direction of that net misstatement hinging on whether the small business threshold is set at $1million or $10million GAR.
Net overstatement is calculated as the gross overstatement (Figure2, Cell C) less the gross understatement (Figure2, CellB).
For banks with $1billion to $10billion in assets, we find a net overstatement of 15percent when using a GAR threshold of
$1million (Table2, panel A). In contrast, at the $10million GAR threshold, we find a net understatement of 23.1percent
(Table2, panel B). However, we know from the previous section that a GAR of $1million is unlikely to be the current maximum
small business size; further, the $10million GAR threshold captures all lending below rather than at that threshold. Therefore, it
is safe to assume that the proxy on net understates lending to small businesses by banks with $1 billion to $10billion in assets.
Table 2
Share of Outstanding C&I Loan Balances
Banks With $1 Billion to $10 Billion in Assets
(Percent Loan-Weighted)
A. $1 Million GAR Threshold
<=$1M >$1M
Small Business
Lending Proxy (A+C)
Metrics
Small Loans (A) 12.9% (C) 24.7% 37.6% p
1
=A/(A+C) 0.343
Large Loans (B) 9.7% (D) 52.7% p
2
=B/(B+D) 0.155
Loans to Small Firms (A+B) 22.6% q=A/(A+B) 0.571
Gross Misstatement (B+C) 34.4%
Net Overstatement (C-B) 15.0% r=p
2
/p
1
0.453
B. $10 Million GAR Threshold
<=$10M >$10M
Small Business
Lending Proxy (A+C)
Metrics
Small Loans (A) 28.3% (C) 7.5% 35.8% p
1
=A/(A+C) 0.791
Large Loans (B) 30.6% (D) 33.6% p
2
=B/(B+D) 0.477
Loans to Small Firms (A+B) 58.9% q=A/(A+B) 0.480
Gross Misstatement (B+C) 38.1%
Net Overstatement (C-B) -23.1% r=p
2
/p
1
0.603
Sources: Small Business Lending Survey (SBLS) Question 14 and fourth quarter 2015 Call Reports.
Note: See Appendix C of the 2018 SBLS report for the survey instrument. See text box below for descriptions and interpretations of the
proxy accuracy metrics presented. SBLS responses are loan-weighted to represent the proportion of U.S. industry lending for banks
with $1 billion to $10 billion in assets.
37
The SBLS also asked questions about new C&I loan originations to businesses of greater than $1million GAR. The results on the mismatch between the
proxy measure and actual lending to businesses using responses to the originations questions are consistent with the results presented here based on
outstanding loan balances.
12
It is also helpful to consider how loans to small businesses are distributed between small and large loans. We denote
the fraction of small business bank lending that is small business loans q = A/(A+B). (Box 1 presents a summary of the
proxy accuracy metrics that we present in this section.) The assumption embedded in the proxy is that all bank loans to
small businesses are small, so that q = 1. This assumption is strongly rejected by the data, regardless of the threshold used;
q = 0.571 at the $1million GAR threshold and q = 0.480 at the $10million GAR threshold. Therefore, nearly half, or possibly
even more than half, of the loans banks make to small businesses are actually large. That small businesses rely more on large
dollar loans than previously assumed carries serious implications for how we view the recovery of small business lending
after the Great Recession, if we rely on proxy data as a measure.
Box 1
Proxy Accuracy Metrics
Metric Description Interpretation
q = A/A+B
Proportion of loans to small businesses that
are below $1 million at origination
q = 0 → All loans to small businesses are large loans
q = 1 → All loans to small businesses are small loans
(proxy assumption)
p
1
= A/A+C
Proportion of small C&I loans (below
$1million at origination) to small businesses
p
1
= 0 → No small loans are made to small businesses
p
1
= 1 → All small loans are made to small businesses
(proxy assumption)
p
2
= B/B+D
Proportion of large C&I loans (above
$1million at origination) to small businesses
p
2
= 0 → No large loans are made to small businesses
(proxy assumption)
p
2
= 1 → All large loans are made to small businesses
r = p
2
/ p
1
Ratio of p
2
to p
1
An inverse measure of validity of proxy’s
assumptions
(Range: 0 to ∞)
r = 0 → p
2
= 0, No large loans are made to small
businesses (proxy assumption)
r = 1 → p
1
= p
2
. An equal proportion of small loans
and large loans are made to small businesses. Proxy
contains no information about small business lending
r to ∞ → p
1
to 0 and p
2
to 1. No small loans are made
to small businesses and all large loans are made to
small businesses
3.2 The Proxy and Small Business Lending Since the Great Recession
The standard narrative since the Great Recession is one of a slow recovery for small business lending, based, in part,
on the slow rebound observed for the proxy. Figures 4A and 4B show how domestic C&I lending in the banking industry has
evolved since 2006 for both the small loans comprising the proxy for small business lending (dashed line) and for C&I loans of
more than $1million (solid line). Figure4A shows the evolution in nominal dollars, which highlights that larger C&I loans are
far greater in magnitude than small C&I loans, while Figure4B shows both series indexed to their own year-end 2008 values,
allowing for an easier comparison of their growth in the pre-recession and post-recession periods. Both large and small loan
volumes grew rapidly in the years before the recession, peaked in 2008 as the recession unfolded, and declined rapidly for
the next several years. Large C&I loans reached their lowest point in 2010, having dropped 21percent from the 2008 peak.
These loans bounced back quickly, however, exceeding their 2008 peak in 2012. By the end of 2018, large C&I loans were
roughly 66percent higher than they were in 2008. By contrast, the proxy reached its lowest point in 2011, 16percent below its
2008 value, but it recovered much more slowly. By the end of 2017, the proxy barely exceeded its 2008 level, and it stood only
6percent higher in nominal terms at the end of 2018 than it did in 2008.
13
Figure 4A
Call Report Outstanding C&I Loan Balances, All Banks
0
500
1,000
1,500
2,000
2006 2009 2012 2015 2018
Billions of Dollars
Small C&I (Proxy Loans of $1M or Less)
Large C&I
Source: Fourth quarter Call Reports.
Nominal Q4 Values
Figure 4B
Call Report Outstanding C&I Loan Balances, All Banks
2018
Index
Small C&I (Proxy Loans of $1M or Less)
Large C&I
Source: Fourth quarter Call Reports.
Nominal Q4 Values (Indexed to 2008=100)
75
100
125
150
175
2006 2009 2012 2015
These trends are often interpreted as indicating that small business lending recovered slowly from the Great Recession;
however, this interpretation relies heavily on the assumption that small businesses receive only small loans from banks, when
we have shown that they draw their bank lending roughly equally in small and large loans, at least for banks with $1 billion to
$10billion in assets. The true evolution of bank lending to small businesses since the crisis therefore likely runs somewhere
between the two lines of Figure4B.
Guided by the results for banks with $1 billion to $10billion in assets, we can calculate how true small business lending
grew under different assumptions. Specifically, we can look at the propensity of small loans to be made to small businesses,
defined as p
1
= A/(A+C), and the propensity of large loans to be made to small businesses, defined as p
2
= B/(B+D),
and use these metrics to derive true small business lending.
38
The assumptions underpinning the proxy are that p
1
= 1 (all
small loans are made to small businesses) and p
2
= 0 (all large loans are made to large businesses). The data reject these
assumptions, regardless of which GAR threshold is used to define small businesses. The point estimate for p
1
is 0.343 using
the $1million GAR threshold, meaning that for every dollar in small C&I lending, businesses under $1million GAR receive
34.3cents (Table2, panel A). Similarly, the point estimate for p
2
is 0.155, indicating that those businesses receive 15.5 cents for
every dollar of large C&I lending.
Therefore, for any given p
1
and p
2
, we can calculate true outstanding small business lending at any point in time by
multiplying the dollar value of total outstanding small C&I loans by p
1
, multiplying the dollar value of total outstanding large
C&I loans by p
2
, and summing the two values.
39
It would, however, be inappropriate to simply plug in the point estimates from
Table2 as they not only represent a fairly small fraction of lending by the banking industry (by banks of $1billion to $10billion
in assets), they also are substantially different based on whether we use the $1million or $10million GAR threshold. For ease
of presentation, we can summarize both p
1
and p
2
into their ratio, r = p
2
/p
1
, and show the evolution of outstanding small
business lending over time when we assume different values for r, relative to a baseline year.
For this metric, when r = 0 (which can only occur when p
2
= 0), no large loans are made to small businesses; small
business lending is then a fixed proportion of small C&I lending, and itspercentage growth over time is the same as that of
38
Note that the data we use here is on outstanding loan balances rather than originations.
39
As discussed in Appendix B, this formula relies on the assumption that p
1
and p
2
do not vary over time, which is likely an oversimplification. It is, however, a
much weaker assumption than what is required to use the proxy, which assumes that p
1
= 1 and p
2
= 0, and that both values do not vary over time.
14
the proxy. On the other hand, r = 1 implies that p
1
= p
2
, so an equal proportion of small and large loans are made to small
businesses, and small business lending is proportionate to total C&I lending. An r > 1 implies that p
1
< p
2
so that a larger
proportion of large loans than small loans are made to small businesses. That scenario is unlikely, however; for banks with
$1billion to $10billion in assets, r is estimated to be 0.453 at the $1million GAR threshold and 0.603 at the $10million GAR
threshold (Table2).
We can therefore apply different values of r to industry-level Call Report data to provide alternate estimates of the
evolution of bank lending to small businesses over time.
40
This allows us to account for the fact that a nontrivial amount of
large C&I loans are directed to small businesses and small C&I loans to large businesses without tying ourselves to a particular
value of r. Figure5 shows the evolution of industry-aggregate outstanding small business lending (including banks of all
sizes that file Call Reports) under three assumptions: r = 0 (equivalent to using the proxy trend), r = 0.1, and r = 0.5, all
indexed to 2008 in the manner of Figure4B. In the more conservative model, where r = 0.1 (which would be consistent with
50percent of small loans and 5percent of large loans being made to small businesses), small business lending still recovers
substantially faster than indicated by the proxy. Under this assumption, small business lending regained its 2008 peak in 2015
instead of 2017, and as of 2018 it would be up 18.9percent from its 2008 level instead of 6.1percent. The more aggressive
model with r = 0.5, which is more in line with what we find in the data, locates the recovery point in 2013 and shows a 2018
level that is 40.7percent higher than the 2008 peak. Overall, these adjustments suggest that small business lending recovered
much more quickly than previously believed.
Figure 5
Small Business Lending Under Dierent Assumptions for r, All Banks
Nominal Q4 Values (Indexed to 2008=100)
75
100
125
150
175
2006 2009 2012 2015 2018
r=0 (Proxy Assumption) r=0.1 r=0.5
Source: Fourth quarter Call Reports.
Note: The value of r represents the share of large C&I outstanding loan balances (>$1 million
at the time of origination) that are to small businesses (p
2
) over small C&I outstanding loan
balances (<=$1 million at the time of origination) that are to small businesses (p
1
), or p
2
/p
1
.
An r of 0 represents the requirements if the proxy were a perfect measure of small business
lending (p
2
=0 and p
1
=1).
Index
40
See Appendix B for details on how we construct our alternate small business lending estimates, including the derivation of how r is a suicient statistic for
p
1
and p
2
.
15
The fact that the overall view of the recovery of small business lending after the Great Recession changes substantially
when we slightly relax the proxy assumptions suggests that researchers should be cautious about relying on the proxy.
Many research papers use either the volume of proxy lending or the ratio of the proxy to total assets as an outcome variable
to measure small business lending. While the inaccuracy of the proxy does not necessarily invalidate the findings of those
papers, they suggest that care should be taken in how these findings are interpreted. Scholars could increase confidence in
their empirical results by adapting the methodology provided in Appendix B into their regression frameworks and testing to
see if their results hold for different values of r.
3.3 Imprecision of the Proxy Across Banks
Despite the net and gross misstatement of small business lending by the proxy, it could still be used to rank banks by
the proportion of their C&I lending that they direct to small businesses, provided that the misstatement was fairly consistent
across all banks. At one extreme, if every bank under- or overstated its small business lending by the same share, then the
proxy could be a valid tool for cross-bank comparisons. Alternatively, the inaccuracy of the proxy as a measure of industry-
level small business lending could be driven by only a few highly inaccurate banks. For example, a small number of banks
may have particular business models that result in their making many small loans to larger businesses. In this case, we would
find that misstatement is close to zero for most banks, with a few highly inaccurate outliers.
We assess the precision of the proxy for measuring small business lending using the distribution of the net misstatement
at the bank level (the share represented by Cell C in Figure2 less the share of Cell B).
41
Figure6 shows the distribution of
net misstatement in terms of thepercentage point (pp) difference between the share of lending measured by the proxy and
thepercentage of C&I lending directed to businesses under the $1million GAR threshold.
If the proxy at the bank level was consistently inaccurate as a measure of small business lending, then we should
observe mass around the industry-level overstatement of 15percent (Table2, panel A) between banks’ share of C&I lending
as measured by the proxy and their share of lending to business with $1million GAR. We find instead that approximately
70percent of banks have overstatement in a broader band between 0 and 30percentage points; approximately 15percent
have overstatement above 30percent. Further, approximately 15percent have a proxy share that understates their
lending to small businesses (defined as $1million GAR). Only about one-fifth (19.8percent) of banks are close to the mean
overstatement of 15percent, with a 10 to 20percentage point overstatement (Figure6).
In addition, the overestimation we observe with the $1million GAR threshold is broad-based and not limited to a few
“bad” or highly divergent “apples.” If this were the case, then we would observe both greater mass around zero (the vertical
line on Figure6) and some mass in the higher overstatement buckets. What we find instead is that only slightly more than
30percent of banks are in the two buckets closest to zero, which rules out the possibility that the proxy overestimation is
driven by a few banks with high overstatement. Further, the median bank is in the 10 to 20percentage point bucket (close to
the mean difference) rather than closer to zero, indicating that the typical bank also exhibits large discrepancies. Overall, we
find wide variation in the difference between lending collected by the proxy and banks’ actual lending to small businesses,
and these differences are not driven by the behavior of a small number of banks.
41
Because this exercise requires a larger sample size to identify the distribution of net misstatement (and not just the mean), we make several choices to
expand our sample and thus keep the number of responses in any category from falling below the allowable reporting threshold for protecting respondents’
identities. This includes exclusively using the $1million GAR threshold (which had a higher response rate than the $10million GAR threshold), including all
banks with more than $1billion in assets, and using broad and irregularly sized buckets. In addition, unlike previous calculations, we use bank-weighting
rather than loan-weighting, since we are assessing precision at the bank level rather than accuracy at the industry level.
16
Figure 6
Distribution of Dierence Between Proxy Lending and Loans to Businesses
With GAR <=$1M as Share of Total C&I Outstanding Balances
Banks With $1 Billion or More in Assets
(Percent Bank-Weighted)
14.0
19.8
28.9
21.6
8.7
6.9
0
5
10
15
20
25
30
35
>30 pp 20 to 30 pp 10 to 20 pp <10 pp <20 pp >20 pp
Net Overstatement Net Understatement
Percent
70.3
15.6
No Dierence Between Proxy
and Loans to Businesses With
GAR <= $1 Million
Sources: Small Business Lending Survey (SBLS) Question 14 and fourth quarter 2015 Call Reports.
Note: See Appendix C of the 2018 SBLS report for the survey instrument. GAR is gross annual revenue.
Broad and irregularly sized buckets for the dierences are used to keep the number of responses in any
bucket from falling below the allowable reporting threshold for protecting respondents’ identities. SBLS
responses are bank-weighted to represent the proportion of banks with assets of $1 billion or more in the
U.S. banking industry that exhibit these characteristics.
Using the lower GAR threshold of $1million, we find that the proxy overstates small business lending for the vast majority
of banks with more than $1billion in assets, mirroring the net overstatement shown in Table2, panel A. However, our intent
is not to suggest that a GAR threshold of $1million is more appropriate; rather, given the data available, that threshold allows
us to document that, across banks, the difference between the lending measured by the proxy and actual lending to small
businesses has a broad and relatively thick-tailed distribution, suggesting an imprecise relationship between the two.
3.4 Ranking Banks Using the Proxy
One consequence of this high dispersion in net misstatement across banks is that the proxy may be unable to
consistently identify which banks are truly doing more small business lending. We investigate this by pairing each bank in our
sample that has $1billion to $10billion in assets with all other similarly sized banks in the sample and comparing their proxy
shares of lending and their actual shares of outstanding C&I lending to businesses with under $1million GAR.
42
For each pair,
we determine both whether the bank with a lower observed proxy share makes a greater proportion of its lending to these
smallest of businesses, and if the bank with a lower proxy share makes a much greater proportion of its lending to these small
businesses (more than a 25percentage point difference).
43
If the proxy were a perfect measure of small business lending (or
if the proxy uniformly misstated actual small business lending for all banks by the same share), we would expect the bank
with the higher proxy share to always make a higher share of actual lending to small businesses than the paired comparison
bank. If the proxy were entirely unrelated to actual small business lending, then banks with a lower proxy share would make a
42
For example, if we had ten banks in our sample, indexed one to ten, we would end up with 45 pairs: bank one and bank two, bank one and bank three, bank
one and bank four, and so on.
43
For each pair, we created an indicator variable for whether the bank in the pair with the lower proxy share made a higher proportion of lending to
businesses with under $1million GAR, and another indicator variable for whether the bank in the pair with the lower proxy share made a much higher (more
than 25percentage points) proportion of their C&I lending to small businesses.
17
higher share of lending to small businesses than paired banks approximately half of the time. Banks with a lower proxy share
of lending make greater shares of actual lending to small businesses in 25.9percent of the bank pairs, and 8.3percent of bank
pairs show more than a 25percentage point difference. (Table3, panel A).
44
These findings are consistent with our earlier
results; the proxy contains information about small business lending but shows limited ability to identify whether one bank is
more likely than another to concentrate its C&I lending to small businesses.
Table 3
Share of Bank Pairs That Exhibit These Characteristics
Banks With $1 Billion to $10 Billion in Assets
(Percent Bank Pair-Weighted)
Mean
Standard
Error
A. All Responding Bank Pairs
Bank With Lower Proxy Has Higher Outstanding Balances to Firms With <$1M GAR 25.9% (0.93)
Bank With Lower Proxy Has >=25pp Higher Outstanding Balances to Firms With <$1M GAR 8.3% (0.59)
B. Only Bank Pairs With >=0.25 Difference in Respective Call Proxies
Bank With Lower Proxy Has Higher Outstanding Balances to Firms With <$1M GAR 17.8% (1.23)
Bank With Lower Proxy Has >=25pp Higher Outstanding Balances to Firms With <$1M GAR 5.4% (0.74)
Sources: Small Business Lending Survey (SBLS) Question 14 and fourth quarter 2015 Call Reports.
Note: GAR is gross annual revenue. See Appendix C of the 2018 SBLS report for the survey instrument. SBLS responses are bank-
weighted to represent the proportion of bank pairs with assets of $1 billion to $10 billion in the U.S. banking industry that exhibit these
characteristics.
Having a much higher proxy lending share than another bank also does not ensure that a bank will have a higher
proportion of its C&I portfolio in small business lending. In 17.8percent of bank pairs in which the difference between the two
banks in their proxy shares is 25percentage points or more, the bank with the lower proxy share has a higher proportion of
small business lending (Table3, panel B). Therefore, the weak correlation between the lending measured by the proxy and
actual lending to small businesses calls into question the reliability of using the proxy to identify which banks are lending a
greater share of their individual C&I portfolio to small businesses, even for much higher proxy share banks.
4 Overall Accuracy of the Proxy: Banks With Less Than $1 Billion in Assets
To assess the overall accuracy of the proxy as a measure of small business lending for the smallest banks, those with
assets of less than $1billion,
45
we rely on a qualitative Yes/No question asked of all banks in the SBLS: “For calendar year 2015,
do you consider largely all of the loans made by your bank for Commercial and Industrial (C&I) purposes to have been made to
borrowers that you consider to be small businesses? ”
46
A majority (83.0percent) of banks with less than $1billion in assets responded affirmatively to the “largely all” survey
question, compared with slightly more than one-third of banks with $1billion to $10billion in assets and about one-tenth
(11.6percent) of the largest banks (Table4). Clearly, smaller banks are more likely than larger banks to focus their C&I lending
activities on small businesses. In the following analysis, we refer to banks that report making largely all of their C&I loans to
small businesses as “small business-focused” (or “focused”) and those that do not as “non-focused.”
44
We conducted a similar exercise using SBLS Question 8 in Appendix C of FDIC (2018) on total new loan originations and Question 10 on new originations to
firms with less than $1million GAR, and found similar results.
45
FDIC (2018) contains a dierent and simpler version of this analysis, which considered only the possibility of gross underestimation, used highly
conservative estimates, and relied on aggregate data rather than microdata.
46
The term “largely all” rather than “all” was used in the survey question given banks’ concerns, shared during field testing interviews, about potentially
misrepresenting themselves to bank regulators by responding airmatively to an all-inclusive statement. See Section II footnote 12 of FDIC (2018).
18
Table 4
Share of Banks Reporting Largely All of Their C&I Loans Made to Borrowers Bank Considers Small
(Percent Bank-Weighted)
Less Than $1B $1B to Less Than $10B $10B or More
83.0% 37.0% 11.6%
Source: Small Business Lending Survey (SBLS) Question 2.
Note: See Appendix C of the 2018 SBLS report for the survey instrument. SBLS responses are bank-weighted to represent the proportion
of banks in the U.S. banking industry that exhibit these characteristics.
If the proxy is an accurate portrayal of small business lending at the bank level, we expect to find that focused banks
have high proxy shares and that non-focused banks have low proxy shares. Banks that report focusing on small business
lending but have low proxy shares are likely making C&I loans to small businesses in ways not fully captured by the proxy,
either because the loans were over $1million at origination or because the bank is making loans for C&I purposes that are not
categorized as C&I loans.
47
The reverse argument applies for non-focused banks with high proxy shares of C&I lending. The
thrust of our analysis is to correct for lending by banks with proxy shares inconsistent with their answers to the qualitative Yes/
No survey question.
We find that banks with less than $1billion in assets generally have a small business lending focus consistent with their
proxy shares, but that a substantial minority of C&I loans are made by focused banks with low proxy shares and non-focused
banks with high proxy shares. More than half (56.2percent) of aggregate C&I lending is by focused banks with high individual
proxy lending shares and 15.5percent by non-focused banks with low proxy shares, for a total of 71.7percent of the market
(Table5). However, 28.3percent of industry-level C&I lending is by banks whose focus on small business lending is not
consistent with their proxy share of lending (18percent by focused banks with low proxy shares and 10.3percent by non-
focused banks with high proxy shares).
Table 5
Share of Industry-Level C&I Outstanding Loan Balances by Whether Banks Focus Their C&I Lending to
Small Businesses and Proxy Share
Banks With Less Than $1 Billion in Assets
(Percent Loan-Weighted)
Proxy Share of Total C&I Lending Focused Non-Focused
Below 50% 18.0% 15.5%
Above 50% 56.2% 10.3%
Total 74.2% 25.7%
Sources: Small Business Lending Survey (SBLS) Question 2 and fourth quarter 2015 Call Reports.
Note: See Appendix C of the 2018 SBLS report for the survey instrument. SBLS responses are loan-weighted to represent the proportion
of U.S. industry lending for banks with less than $1 billion in assets.
Our approach to determining misstatement among the smallest banks is to (1) identify banks with mismatched survey-
reported focus and Call Report proxy shares, (2) reasonably adjust their proxy shares to match their focus, (3) recalculate small
business lending at the industry level, and (4) compare this recalculated small business lending to the original proxy.
48
We
adjust the proxy share by setting a ceiling for non-focused banks and a floor for focused bank: proxy shares of non-focused
banks are downwardly adjusted if the share is above the ceiling, and proxy shares of focused banks are upwardly adjusted if
47
See Section II of FDIC (2018) for a more detailed discussion of this issue.
48
See Appendix C for a technical description of the adjustment process.
19
the share is below the floor. This approach reduces the small business lending by non-focused banks with implausibly high
proxy shares (such as a bank that does not consider “largely all” of its C&I lending to be to small businesses and yet has a
proxy share of 90percent) and increases the small business lending by focused banks with implausibly low proxy shares (such
as a bank that considers “largely all” its lending to be to small businesses and yet has a proxy share of 25percent). Since there
are multiple plausible interpretations of “largely all,” we use several floor and ceiling assumptions.
For example, suppose we used a floor of 60percent and a ceiling of 40percent. For the non-focused bank with a proxy
share of 90percent, we would downward-adjust their share of proxy lending to the 40percent ceiling. The proxy lending share
for a non-focused bank with a proxy share of 25percent would not be adjusted, as their share is already below the ceiling.
Similarly, proxy lending share for a focused bank with a proxy share of 90percent would not be adjusted, as it is above the floor.
But proxy lending share for a focused bank with a proxy share of 25percent would be upwardly adjusted to the 60percent floor.
Under a range of reasonable assumptions, we find evidence that the proxy mildly understates true small business
lending by banks with less than $1billion in assets. Figure7 shows estimates and 95percent confidence intervals of the small
business share of C&I lending of these-sized banks at the industry level, using different floor and ceiling assumptions, as well
as the unaltered proxy share on the left. The estimates range from 60.5percent to 72.8percent, with the lower bounds of
their confidence intervals mostly above the proxy share of 60.4percent. This suggests that the proxy understates rather than
overstates small business lending by the smallest banks. Under a moderate set of assumptions (a ceiling of 40percent small
business lending for non-focused banks and a floor of 70percent for focused banks), we find an estimate of 65.9percent,
5.5percentage points greater than the proxy share.
Figure 7
Estimates of the Share of C&I Lending to Small Businesses by Banks With
Less Than $1 Billion in Assets in Fourth Quarter 2015
(Percent Loan-Weighted)
Percent
60.5
60.4
65.9
72.8
0
10
20
30
40
50
60
70
80
90
100
Floor: 60 Floor: 70 Floor: 80
Ceiling: 30 Ceiling: 40 Ceiling: 50 Ceiling: 60
Proxy Share
Sources: Small Business Lending Survey (SBLS) Questions 2 and 4 and fourth quarter 2015 Call Reports.
Note: See Appendix C of the 2018 SBLS report for the survey instrument. A “floor” is applied to adjust
upward the proxy share of banks that respond “Yes” to the “largely all” question (focused banks) but have
proxy shares below the indicated minimum floor. A “ceiling” is applied to adjust downward the proxy
share of banks that respond “No” (non-focused banks) but have proxy shares above the indicated
maximum ceiling. SBLS responses are loan-weighted to represent the proportion of U.S. industry lending
for banks with less than $1 billion in assets.
20
5 Conclusions
In this study, we use novel data from the FDIC’s Small Business Lending Survey to assess how well C&I loans of under
$1million at origination proxy for lending to small businesses. By using questions that provide a coarse disaggregation of
lending by loan and firm size, we are able to investigate the accuracy of the proxy for banks with $1billion to $10billion in
assets, the implications of inaccuracies for interpreting overall small business lending trends, the dispersion of bank-level
differences between the observed proxy and actual small business lending, and whether the proxy can reliably identify
whether one bank has a greater share of its C&I portfolio in small business lending relative to another.
We investigated to what extent the basic assumptions behind the proxy— small loans are extended to small firms and
large loans are extended to large firms—hold. We find that these two assumptions are not well-supported by the analysis.
For banks with $1billion to $10billion in assets, large shares of small business lending are grossly misstated if relying on the
proxy. This is true using either of the small business size thresholds available in the survey data. More than 30percent of C&I
lending by banks with $1billion to $10billion in assets are small loans to large firms or large loans to small firms.
The direction of overall net misstatement depends critically on the maximum small business size assumed. For banks
with assets of $1billion to $10billion, the proxy, on net, substantially overestimates small business lending if relying on a
GAR of less than $1million to identify small businesses, but it substantially underestimates small business lending if using a
GAR of less than $10million. Given that a GAR of $10million likely captures more true small business lending than a GAR of
$1million, the proxy likely net understates these banks’ lending to U.S. small businesses by up to 23percent, as a share of the
banks’ aggregate industry-level C&I lending.
Limitations of the proxy for measuring small business lending have repercussions for an understanding of small business
lending trends. The common narrative of small business lending after the Great Recession highlights its slow recovery, based
in part on the recovery observed if relying on the proxy. However, if we weaken the assumptions underlying the proxy to
match banks’ actual small business lending activity observed from the survey data (where some small loans go to larger firms
and some large loans to go to small firms), small business lending by the banking industry recovers much faster, reaching its
previous high two to four years earlier than previously believed.
Further, we find that banks exhibit wide dispersion in how much the proxy over- or understates their small business
lending. This variation raises questions about the ability of the proxy to correctly rank banks in terms of the shares of their
C&Iportfolio devoted to small business lending; approximately one-quarter of the time when two banks are compared, the
bank with the lower proxy share actually makes a larger proportion of its lending to small businesses, as measured by the
survey data.
Taken together, these findings suggest that the proxy—while reliably collecting data on small loans to businesses—should
be used with caution. When relying on the proxy to measure small business lending, alternate data sources should also be
used and robustness tests under different assumptions should be implemented.
Finally, for the smallest banks, those with less than $1billion in assets, we conduct a simple exercise to assess the overall
accuracy of the proxy for measuring small business lending by these banks. We rely on a qualitative question on whether
the bank focuses its C&I lending to small businesses, combined with some reasonable assumptions, to either upwardly or
downwardly adjust the bank’s proxy share of industry-level C&I loans. We find that the proxy likely only mildly understates
small business lending for the smallest banks.
21
References
“A Guide to CRA Data Collection and Reporting.” Federal Financial Institutions Examination Council.
https://www.iec.gov/cra/guide.htm.
Berger, Allen N., Nathan H. Miller, Mitchell A. Petersen, Raghuram G. Rajan, and Jeremy C. Stein. “Does Function Follow
Organizational Form? Evidence From the Lending Practices of Large and Small Banks.” Journal of Financial Economics 76,
no. 2 (2005): 237–69. https://doi.org/10.1016/j.jfineco.2004.06.003.
Berger, Allen N., and Gregory F. Udell. “Small Business Credit Availability and Relationship Lending: The Importance of Bank
Organisational Structure.” The Economic Journal 112, no. 477 (February 2002): F32–F53.
https://doi.org/10.1111/1468-0297.00682.
Bordo, Michael, and John Duca. 2018. “The Impact of the Dodd-Frank Act on Small Business.” National Bureau of Economic
Research Working Paper 24501.
Cole, Rebel. 2018. “How Did Bank Lending to Small Business in the United States Fare Aer the Financial Crisis?” U.S. Small
Business Administration Research Study 438.
“Community Banks: Eect of Regulations on Small Business Lending and Institutions Appears Modest, but Lending Data
Could Be Improved.” U.S. Government Accountability Oice, September 5, 2018.
https://www.gao.gov/reports/GAO-18-312/.
Cortés, Kristle, Yuliya Demyanyk, Lei Li, Elena Loutskina, and Philip E. Strahan. “Stress Tests and Small Business Lending.”
Journal of Financial Economics 136, no.1 (2020): 260-279. https://doi.org/10.1016/ j.jfineco.2019.08.008.
Dilger, Robert Jay. “Small Business Size Standards: A Historical Analysis of Contemporary Issues.” Congressional Research
Services (February 1, 2019).
Federal Deposit Insurance Corporation. 2018. FDIC Small Business Lending Survey 2018.
https://www.fdic.gov/bank/historical/sbls/.
Horvitz, Daniel G., and Donovan J. Thompson. “A Generalization of Sampling Without Replacement From a Finite Universe.”
Journal of the American Statistical Association 47.260 (1952): 663–685.
Reichow, Kelsey. 2017. “Small-Business Lending Languishes as Community Banking Weakens.” Federal Reserve Bank of Dallas
Economic Letter 12, no. 3.
Robb, Alicia M., and David T. Robinson. 2010. “The Capital Structure Decisions of New Firms.” National Bureau of Economic
Research Working Paper 16272.
Udell, Gregory F. “What’s in a Relationship? The Case of Commercial Lending.” Business Horizons 51, no. 2 (2008): 93–103.
https://doi.org/10.1016/j.bushor.2007.10.005.
22
Appendix A. Weighting the Small Business Lending Survey
In this report, we are typically interested in estimating the proportion of C&I loans that are made to small or large firms,
at an industry or sub-industry level. To generate these estimates using data from the SBLS, we adjust the standard Horvitz-
Thompson estimator used for disproportionate stratified random samples. This estimator uses inverse probability weights
to estimate the average proportions at the bank level; we further weight banks by their outstanding C&I lending to generate
proportions at the industry level. This appendix derives the estimator used in most of the report and shows that it produces
unbiased estimates.
Suppose we wished to determine the proportion of all C&I bank loans which are both small and that are made to small
firms, which we denote a
agg
. If we had data from every bank, we would calculate this proportion by summing up the small
C&Iloans to small firms at each bank and dividing by the sum of total C&I loans at each bank:
a
agg
=
∑
i∈I
A
i
∑
i∈I
Z
i
(1)
where A
i
denotes the dollar value of outstanding small C&I loans to small firms, Z
i
denotes total outstanding C&I loans for
bank i, and I is the entire population of banks. If we denote the bank-level proportion to be a
i
=
A
i
Z
i
and denote
=
∑
i∈I
Z
i
Z
tobe total aggregate outstanding C&I lending, we can rewrite this as:
a
agg
=
1
∑
i∈I
Z
i
a
i
Z
(2)
This is simply a weighted average of a
i
, weighing each observed bank by it’s total outstanding C&I lending.
In addition to this loan-weighting, we must also account for the fact that the SBLS consists of a stratified random sample
and not a full census, and that we will be constructing an estimate â
agg
of a
agg
rather than reporting the value directly.
1
Since
eligible banks in different strata have different probabilities of selection, to produce an unbiased estimator we must include
an inverse-probability weight (as in the commonly used Horvitz-Thompson estimator) in addition to the loan-weight. The
estimator weuse is:
Z
ˆa
agg
=
1
∑
j
J
1
π
j
Z
j
a
j
∈
(3)
where J is the set of banks included in the survey and π 
j
is the probability that bank j was ultimately included in the survey.
Per Horvitz and Thompson (1952), the expectation of â
agg
is equal to a
agg
:
E[ ˆa
agg
]=E[]=
∑
iI
Z
i
a
i
=a
agg
Z
1
∑
j
J
1
π
j
Z
j
a
j
∈
∈
Z
1
(4)
making â
agg
an unbiased estimator of the a
agg
. We use similar estimators for the B, C, and D cells of Table 2.
1
See FDIC (2018) Appendix A for more details on the survey methodology, including strata defininitions and sample sizes.
23
Appendix B. Derivation of Small Business Lending Over Time Under
Alternate Assumptions
This appendix describes how we calculate the time series of small business lending (indexed to 2008) under alternate
assumptions from those embedded in the small business lending proxy, as discussed in Section 3. Instead of assuming
that all small loans are made to small firms and all large loans are made to large firms, we assume that, across the entire
banking industry, both a fixed proportion of small loans p
1
[0,1] go to small businesses and a fixed proportion of large
loans p
2
[0,1] go to small businesses. We must still make the strong assumption that p
1
and p
2
are time-invariant.
However, this is a much weaker set of assumptions than the underlying proxy assumption p
1
= 1 and p
2
= 0, as the proxy
assumption implictly embeds the time-invariance assumption as well.
2
We rely solely on data from the Call Report to perform
this exercise rather than supplementing with SBLS data, although our chosen parameter values are informed by results from
the SBLS.
We denote observed annual aggregate small C&I loans in the Call Report (i.e. the lending included in the proxy) in year
t as X
t
and observed annual aggregate large C&I loans as Y
t
. These values are drawn from the fourth quarter Call Report
of each year, as reported in the FDIC’s Quarterly Banking Profile.
3
For a given set of parameters p
1
and p
2
, we can calculate
annual aggregate lending to small firms S
t
from X
t
and Y
t
:
S
t
(
p
1
,p
2
)
=p
1
X
t
+p
2
Y
t
(5)
where (as above) p
1
is the fraction of small loans made to small firms and p
2
is the fraction of large loans made to large firms,
both of which we assume to be constant over time.
Because we are interested in the nominal growth of small business lending over time rather than its dollar value, we
index S
t
to the year 2008, denoting this indexed value M
t
. Using M
t
instead of S
t
will allow us to reduce the parameter space
by using a single parameter r =
p
2
p
1
instead of assigning p
1
and p
2
independently. The definition of M
t
is thus
M
t
S
t
( p
1
,p
2
)
S
2008
(
p
1
,p
2
)
(p
1
,p
2
) =
(6)
Substituting the definition of S
t
into this equation, we find
M
t
( p
1
,p
2
) =
p
1
X
t
+p
2
Y
t
p
1
X
2008
+p
2
Y
2008
(7)
M
t
( p
1
,p
2
) =
X
t
+
p
2
p
1
Y
t
X
2008
+
p
2
p
1
Y
2008
(8)
We can then express M
t
solely as a function of the parameter r defined above:
M
t
(r) =
X
t
+rY
t
X
2008
+rY
2008
(9)
Note that we must maintain the assumption that p
1
and p
2
are time-invariant (and not just that r is time-invariant), even
though they do not appear in the final equation. Since X
t
and Y
t
are in the Call Report, we can therefore construct and graph
M
t
at various levels of r. These time series are plotted for r = 0.1 and r = 0.5 in Figure 5.
2
Technically, for the M
t
index we discuss later, the small business lending proxy assumption is somewhat weaker; specifically, that p
1
> 0, p
2
= 0, and that
p
1
is time-invariant. As long as no large loans go to small firms and a fixed proportion of small loans go to small firms, it makes no dierence what that fixed
proportion is for our understanding of small business lending in 2018 relative to small business lending in 2008.
3
For small C&I loans, we use the data series for outstanding C&I loans of under $1 million at origination. For large C&I loans, we use the dierence between
outstanding domestic C&I loans to U.S. addresses and small C&I loans. All values are drawn from the Call Report data used in the third quarter 2019 Quarterly
Banking Profile.
24
Appendix C. Estimating Small Business Lending by Banks With Under $1 Billion
in Assets
This appendix describes how we estimate small business lending by banks with under $1 billion in assets using the proxy
data from the Call Report and data from the SBLS, as discussed in Section 4. In this analysis, we assume that, at least for these
banks, the proxy is a generally accurate measure of the proportion of C&I loans that go to small firms, except when banks
explicitly contradict this assumption in their responses to the SBLS.
First, we estimate small business lending by each individual bank by taking the proxy and, if necessary, adjusting it to
match whether the bank reported a focus on small business lending in the SBLS. This adjustment is made by setting a ceiling
for non-focused banks and a floor for focused banks; non-focused banks with a small business lending proxy above the
ceiling are downward-adjusted to the ceiling, while focused banks with a small business lending proxy below the floor are
upward-adjusted to the floor. These individual-level estimates are then weighted by the bank’s sampling weight multiplied by
total C&I lending to produce an industrywide estimate. These industry-wide estimates are reported, for a variety of ‘floor’ and
‘ceiling’ values, in Figure 7.
Formally, we define the following variables for each bank i based on data from the Call Report and the SBLS:
x
i
[0,1] is the proportion of outstanding C&I loans that were $1 million or less at origination (i.e., the proxy share)
observed in the Q4 2015 Call Report.
f
i
{0,1} is the response to Question 2 of the SBLS (“For calendar year 2015, do you consider largely all of the loans
made by your bank for Commercial and Industrial (C&I) purposes to have been made to borrowers that you consider to be small
businesses?”), with f
i
= 0 indicating “no” and f
i
= 1 indicating “yes.” When f
i
= 1 a bank is referred to as “focused” on small
business lending and when f
i
= 0 it is referred to as “non-focused.”
Finally, there are two parameters that are not drawn from the data: the ceiling value t
c
[0,1] (the maximum proportion
of small business lending that we allow for non-focused banks) and the floor value t
f
[0,1] (the minimum proportion
of small business lending that we allow for focused banks). We estimate small business lending under a range of different
assumptions for these parameters, based on how banks might reasonably interpret the meaning of the term “largely all” as
used in the survey question.
The definition for small business lending
ˆ
p
i
is then
ˆ
p
i
=
t
f
if f
i
=
1 and
x
i
<t
f
t
c
if f
i
=0and x
i
>t
c
x
i
otherwise
(10)
Appendix Figure 1 gives a visual example of the adjustments that could occur in a hypothetical dataset, for the specific
case when we set t
f
= 0.7 and t
c
= 0.4. Each point in the figure represents a bank with q
i
on the horizontal axis and the
estimated small business lending
ˆ
p
i
assigned to it on the vertical axis. Banks that did not focus on small business lending
( f = 0, the blue dots) and had a proxy share under 0.4 were unadjusted; those with a proxy share above 0.4 were adjusted
down to that ceiling. Banks that did focus on small business lending ( f = 1, the orange triangles) and had a proxy share
under 0.7 were adjusted up to the 0.7 floor; those with a proxy share above 0.7 were left unadjusted. For example, bank a is
non-focused ( f
a
= 0 ) and has a proxy share q
a
= 0.35, which is below the t
c
threshold, and so is estimated to have a small
business lending share
ˆ
p
a
= q
a
= 0.35. Bank b is also non-focused ( f
b
= 0 ) but has a proxy share q
b
= 0.6, above the t
c
threshold, so its estimated small business lending share is
ˆ
p
b
= t
c
= 0.4.
25
After constructing the bank-level
ˆ
p
i
, we aggregate these values into an industry-wide estimate for banks with under
$1billion in assets using a loan-weighted average (as described in Appendix A):
Z
ˆ
p
agg
=
1
∑
j
J
1
π
j
Z
jj
∈
ˆ
p
(11)
Z
ˆ
P
agg
=
ˆ
p
agg
(12)
where, as in Appendix A, π
j
is the probability that bank j was ultimately included in the survey and
Z
is total aggregate
outstanding C&I lending.
We calculate
ˆ
P
agg
based on a range of assumptions for t
f
and t
c
. Figure 7 shows the point estimates and 95 percent
confidence intervals for
ˆ
P
agg
when t
f
{0.6,0.7,0.8} and t
c
{0.3,0.4,0.5,0.6}.
Appendix Figure 1
Example Adjustments to Proxy Share
Source: Author’s calculations.
Estimated Small Business
Lending Share (p)
ˆ
0
0.4
0.2
0.6
0.8
1.0
Proxy Share (x)
0 0.1 0.4 0.6 0.8 1.0
Not Focused on Small Business Lending (f = 0)
tc = 0.4
Focused on Small Business Lending (f = 1)
tf = 0.7