Bank Liquidity Creation

Allen Berger
Christa Bouwman

Although the modern theory of financial intermediation portrays liquidity creation as an

essential role of banks, comprehensive measures of bank liquidity creation do not exist. We
construct four measures and apply them to data on virtually all U.S. banks from 1993 to
2003. We find that bank liquidity creation increased every year and exceeded $2.8 trillion in
2003. Large banks, multibank holding company members, retail banks, and recently merged
banks created the most liquidity. Bank liquidity creation is positively correlated with bank
value. Testing recent theories of the relationship between capital and liquidity creation, we
find that the relationship is positive for large banks and negative for small banks. (JEL G21,
G28, G32)

According to the modern theory of financial intermediation, banks exist be- cause they
perform two central roles in the economy—they create liquidity and they transform risk.1
Analyses of banks’ role in creating liquidity and thereby spurring economic growth have a
long tradition, dating back to Adam Smith (1776).2 Modern reincarnations of the idea that
liquidity creation is central to banking appear most prominently in the formal analyses in
Bryant (1980) and

Diamond and Dybvig (1983). These theories argue that banks create liquidity on the balance
sheet by financing relatively illiquid assets with relatively liquid liabilities. Holmstrom and
Tirole (1998) and Kashyap, Rajan, and Stein (2002) suggest that banks also create liquidity
off the balance sheet through loan commitments and similar claims to liquid funds.3

Banks’ role as risk transformers is also well studied. A vast literature has emerged on bank
risk taking and prudential regulation, supervision, and market discipline to control risk-
taking behavior. According to the risk transformation theories, banks transform risk by
issuing riskless deposits to finance risky loans (e.g., Diamond 1984; Ramakrishnan and
Thakor 1984; Boyd and Prescott 1986). Risk transformation may coincide with liquidity
creation, as for example, when banks issue riskless liquid deposits to finance risky illiquid
loans. However, liquidity creation and risk transformation do not move in perfect tandem—
the amount of liquidity created may vary considerably for a given amount of risk
transformed. It is therefore essential to study both roles of banks and to distinguish between
them.

Most of the empirical literature has focused on banks’ role as risk transformers, rather than on
their role as liquidity creators. Consequently, comprehensive empirical measures of bank
liquidity creation are conspicuously absent, making it difficult to address numerous questions
of research and policy interest. How much liquidity does the banking sector create? How has
bank liquidity creation changed over time? How does it vary in the cross section? Which banks
create the most and least liquidity? What are the value implications of bank liquidity
creation? Moreover, without measures of liquidity creation in hand, it is not possible to
examine policy-relevant issues, such as the effect of bank capital on bank liquidity creation.

Our main goals here are threefold. Our first goal is to develop comprehensive measures of bank
liquidity creation. We create four such measures that differ in how off-balance sheet activities
are treated and how loans are classified. Our second goal is to use these measures to gain a
deeper insight into banks’ role as liquidity creators by addressing the questions highlighted
above. Specifically, we explore how much liquidity banks create, how liquidity creation has
changed over time, how it varies in the cross section, which banks create the most and least
liquidity, and how liquidity creation is related to bank value. We do this by applying our
liquidity creation measures to data on virtually all U.S. banks over 1993–2003, by splitting
the data in various ways (by bank size, bank holding company status, wholesale versus retail
orientation, and merger status), by contrasting the top 25% and bottom 25% of liquidity creators
in each size class, and by examining correlations between liquidity creation and bank value.
The current financial crisis raises the additional question of how bank liquidity creation
responds during crises. This is one of the issues addressed in a follow-up paper that uses one
of the measures of liquidity creation developed here (Berger and Bouwman 2008). Our third
goal is to use our liquidity creation measures to examine the policy-relevant issue mentioned
above—the effect of bank capital on bank liquidity creation. Some recent theories predict that
bank capital reduces bank liquidity creation, while others predict that capital makes banks
capable of absorbing more risk, and thereby allows them to create more liquidity. We develop
economic intuition about the types of banks for which these opposing effects may dominate,
and test the relationship between capital and liquidity creation predicted by the theories.

To construct our liquidity creation measures, we use a three-step procedure. In step 1, we

classify all bank assets, liabilities, equity, and off-balance sheet activities as liquid,
semiliquid, or illiquid. We do this based on the ease, cost, and time for customers to obtain
liquid funds from the bank, and the ease, cost, and time for banks to dispose of their
obligations in order to meet these liquidity demands. Our use of just three liquidity
classifications (liquid, semiliquid, and illiquid) is a necessary simplification—any finer
distinctions would have to be made rather arbitrarily. In step 2, we assign weights to the
activities classified in step 1. The weights are consistent with the theory—maximum (i.e.,
dollar-for-dollar) liquidity is created when illiquid assets are transformed into liquid liabilities
and maximum liquidity is destroyed when liquid assets are transformed into illiquid
liabilities or equity. In step 3, we construct four liquidity creation measures by combining
the activities as classified in step 1 and as weighted in step 2 in different ways. The measures
classify all activities other than loans by both product category and maturity but—due to
data limitations—classify loans based either solely on category (“cat”) or solely on maturity
(“mat”). To assess how much liquidity banks create on the balance sheet versus off the
balance sheet, we alternatively include off-balance sheet activities (“fat”) or exclude them
(“nonfat”). We thus construct liquidity creation measures based on the four combinations:
“cat fat,” “mat fat,” “cat nonfat,” and “mat nonfat.” As explained below, “cat fat” is our
preferred measure.

When we apply our measures to the data, we find that the U.S. banking industry created
$2.843 trillion in liquidity in 2003 using our preferred “cat fat” measure.4 This equals 39%
of bank gross total assets or GTA (total assets plus allowance for loan and lease losses and the
allocated transfer risk reserve) and is 4.56 times the overall level of bank equity capital,
suggesting that the industry creates $4.56 of liquidity per $1 of capital. To provide further
perspective on liquidity creation relative to bank size, note that bank liquidity creation equals
70% of gross loans and 58% of total deposits.

Liquidity creation has grown dramatically over time: it increased every year and virtually
doubled between 1993 and 2003 based on our preferred “cat fat” measure. This evidence
contradicts the notion that the role of banks in creating liquidity has declined due to the
development of capital markets. Our results are fairly similar when we calculate liquidity
creation using our “mat fat” measure, which classifies loans based on maturity instead of
category. Results based on our “nonfat” measures reveal that the banking sector only creates
about half of its liquidity on the balance sheet, highlighting the importance of liquidity
created off the balance sheet as in Holmstrom and Tirole (1998) and Kashyap, Rajan, and
Stein (2002).
Liquidity creation differs considerably among large banks (GTA exceeding $3 billion),
medium banks (GTA $1 billion–$3 billion), and small banks (GTA up to $1 billion)
(measured in real 2003 dollars). We split our sample by bank size because size differences
among banks are substantial and various empirical studies have shown that components of
liquidity creation vary greatly by bank size. Based on our preferred “cat fat” measure, large
banks are responsible for 81% of industry liquidity creation, while comprising only 2% of
the sample observations. All size classes generate substantial portions of their liquidity off
the balance sheet, but the fraction is much higher for large banks. All size classes
increased liquidity creation in real terms over the sample period. While large banks showed
the greatest growth in the dollar value of liquidity creation, small banks had the greatest
growth in liquidity creation divided by GTA, equity, loans, and deposits.

Liquidity creation also varies with several key bank characteristics. It is starkly different for
banks split by bank holding company status, wholesale versus retail orientation, and merger
status. Based on our preferred “cat fat” measure, banks that are members of a multibank
holding company, have a retail orientation, and engaged in M&A activity during the prior
three years created most of the banking industry’s overall liquidity. These banks also show
the strongest growth in liquidity creation over time.

Liquidity creation is also positively linked with value. We examine the value implications of
liquidity creation by focusing on listed independent banks and banks that are part of listed
bank holding companies. We find that banks and bank holding companies that create more
liquidity have significantly higher market-to-book and price-earnings ratios to create
liquidity. Liquidity creation exposes banks to risk—the greater the liquidity created, the
greater are the likelihood and severity of losses associated with having to dispose of illiquid
assets to meet customers’ liquidity demands (Allen and Santomero 1998; Allen and Gale
2004). Capital absorbs risk and expands banks’ risk-bearing capacity (e.g., Bhattacharya and
Thakor 1993; Repullo 2004; Von Thadden 2004; Coval and Thakor 2005), so higher capital
ratios may allow banks to create more liquidity. We refer to this second set of theories
collectively as the “risk absorption” hypothesis, while recognizing that the theories together
rather than separately produce this prediction.

Both the “financial fragility-crowding out” and the “risk absorption” effects may apply in
differing degrees to liquidity creation by different banks, so the relevant empirical issue is
discovering the circumstances under which each effect empirically dominates. We address
this by testing whether the net effect of bank capital on liquidity creation is negative or positive
for different sizes of banks. We expect that the “financial fragility-crowding out” effect is likely
to be relatively strong for small banks. One reason is that small banks deal more with
entrepreneurial-type small businesses, where the close monitoring highlighted in Diamond
and Rajan (2000, 2001) is important. A second reason is that small banks tend to raise funds
locally, so that capital may “crowd out” deposits as in Gorton and Winton (2000).5 This
effect is likely to be relatively weak for large banks that can more easily access funding from
national or international capital markets. In contrast, the “risk absorption” effect is likely to be
stronger for large banks because they are generally subject to greater regulatory scrutiny and
market discipline than small banks, which may affect their capacity to absorb risk. Since
medium banks fall somewhere in the middle, we expect that either effect may dominate for
these banks or that these effects may simply offset each other.

While the theories suggest a causal relationship from capital to liquidity creation, in practice
both may be jointly determined. This makes it challenging to establish causation. Although
we present regression results using lagged capital to mitigate the endogeneity problem, we
interpret our results with care and do not claim to have established causation. Nonetheless, at
a minimum, the results should be viewed as interesting correlations between capital and
liquidity creation that are consistent with the theories. We also address the endogeneity
problem more directly with instrumental variable regressions and obtain consistent results.

Keeping these precautions in mind, we test the relationship between capital and liquidity
creation by regressing the dollar amount of bank liquidity creation (calculated using our four
measures and normalized by GTA) for each bank- year observation on the bank’s lagged
equity capital ratio and a number of control variables. Since the effect of capital on liquidity
creation may be driven by banks’ role as risk transformers rather than their role as liquidity
creators, all our regressions control for bank risk. We use three-year lagged average values of
capital and the other exogenous variables to mitigate potential endogeneity problems, as
lagged values represent earlier bank decisions. We run the tests separately for large,
medium, and small banks to allow for the possibility that capital may affect these banks
differently.

We find empirical support for both hypotheses. For large banks, the relationship between
capital and liquidity creation is positive, consistent with the expected empirical dominance of
the “risk absorption” effect. In sharp contrast, for small banks, the relationship between
capital and liquidity creation is negative, consistent with the expected dominance of the
“financial fragility- crowding out” effect for these institutions. The relationship is not
significant for medium banks, suggesting that the two effects cancel each other out for this
size class. To understand more deeply why these differences exist, we also examine the
relationship between capital and the individual components of liquidity creation and find
substantial differences across size classes in which components are significantly correlated
with capital.

We test the robustness of our main regression results in various ways. First, our main
liquidity creation measures are based on the ease, cost, and time for customers to obtain
liquid funds, and the ease, cost, and time for banks to dis- pose of obligations to provide
liquid funds. Since buyers of loan commitments and letters of credit may not fully draw
down committed funds, we construct an alternative measure that incorporates the likelihood
with which these customers request actual funds. Second, we construct a liquidity creation
measure that uses an alternative way to establish which assets are securitizable. Third, to
address a potential concern that our dependent variable (liquidity creation) includes current
−
equity capital with a weight of ½ while our key independent variable is the lagged equity capital
ratio, we construct an alternative liquidity creation measure that does not include current
capital. Fourth, since the theories sometimes use a broader definition of equity that includes
other funds that cannot easily run on banks, we use an alternative capital ratio that includes
equity plus other financial instruments, such as long-term subordinated debt (total capital
specified in the Basel I capital standards). Fifth, because the intertemporal and cross-
sectional liquidity creation patterns are so different for banks split by holding company
status, wholesale versus retail orientation, and merger status, we rerun our regressions for
these subsamples. The results of all of these robustness checks reinforce our main findings
and suggest that the relationship between capital and liquidity creation is positive for large
banks, insignificant for medium banks, and negative for small banks.

Since capital and liquidity creation may be jointly determined and the use fraction of seniors
in the markets in which a bank operates as potentially valid instruments for lagged
capital. Based on first-stage regression results, we use the tax rate as an instrument for large
banks only and the fraction of seniors as an instrument for small banks only. Using these
instruments, we find—consistent with our main findings—that the effect of capital on
liquidity creation is significantly positive for large banks and significantly negative for small
banks.

The remainder of the article is organized as follows. Section 1 reviews the literature.
Section 2 describes the construction of our liquidity creation measures. Section 3 discusses
our panel dataset of U.S. banks over 1993–2003, shows how bank liquidity creation varies
over time and in the cross section, contrasts high and low liquidity creators, and explores the
 value implications of bank liquidity creation. Section 4 outlines our regression framework,
and Section 5 contains our regression results. Section 6 addresses robustness issues, and
Section 7 concludes.

1. Literature Review

In this section, we provide a brief review of the literature to place our article in context. Our
research is related to three strands of literature: the measurement of bank liquidity creation;
the theories of the effect of capital on liquidity creation; and the empirical studies of capital and
liquidity creation. We discuss these three literatures in turn.

1.1 Measurement of bank liquidity creation

We are aware of only one paper that attempts to measure bank liquidity creation. Deep and
Schaefer (2004) construct a measure of liquidity transformation and apply it to data on the two
hundred largest U.S. banks from 1997 to 2001. They define the liquidity transformation gap or
−
“LT gap” as (liquid liabilities liquid assets)/total assets. They consider all loans with maturity
of one year or less to be liquid, and they explicitly exclude loan commitments and other off-
balance sheet activities because of their contingent nature. They find that the LT gap is about
20% of total assets on average for their sample of large banks. The authors conclude that
these banks do not appear to create much liquidity, and run some tests to explain this finding,
examining the roles of insured deposits, credit risk, and loan commitments.

The LT gap is an intuitive step forward, but we do not believe that it is sufficiently
comprehensive. To illustrate, we highlight a few key differences between their approach and
ours. First, we include virtually all commercial banks and compare findings for large,
medium, and small banks, rather than including only the largest institutions. Second, our
preferred “cat fat” liquidity creation measure classifies loans by category (“cat”), rather than
by maturity. We treat business loans as illiquid regardless of their maturity because banks
generally cannot easily dispose of them to meet liquidity needs, but we treat residential
mortgages and consumer loans as semiliquid because these loans can often be securitized
and sold to meet demands for liquid funds. Third, our preferred measure includes off-
balance sheet activities (“fat”), consistent with the arguments in Holmstrom and Tirole
(1998) and Kashyap, Rajan, and Stein (2002). In our less-preferred liquidity measures, we
classify loans by maturity (“mat”) and exclude off-balance sheet activities (“nonfat”) to
deter- mine the effects of these assumptions. As discussed in Section 2, the LT gap is
conceptually close to our “mat nonfat” measure.

1.2 Bank liquidity creation and capital: the theories

In the seminal theories on financial intermediary existence highlighted above, banks do not
hold any capital. Bank capital was introduced in subsequent papers, which argue that the
primary reason why banks hold capital is to absorb risk, including the risk of liquidity
crunches, protection against bank runs, and various other risks, most importantly credit risk.
Although the reason why banks hold capital is motivated by their risk transformation role, recent
theories suggest that bank capital may also affect banks’ ability to create liquidity. These
theories produce opposing predictions on the link between capital and liquidity creation.

One set of theories, which we refer to collectively as the “financial fragility- crowding out”
hypothesis, predicts that higher capital reduces liquidity creation. Diamond and Rajan (2000,
2001) focus on financial fragility. They model a relationship bank that raises funds from
investors to provide financing to an entrepreneur. The entrepreneur may withhold effort,
which reduces the amount of bank financing attainable. More importantly, the bank may also
withhold effort, which limits the bank’s ability to raise financing. A deposit contract
mitigates the bank’s holdup problem, because depositors can run on the bank if the bank
threatens to withhold effort, and therefore maximizes liquidity creation. Providers of capital
cannot run on the bank, which limits their willingness to provide funds, and hence reduces
liquidity creation. Thus, the higher a bank’s capital ratio, the less liquidity it will create.6
Note that the negative effect of capital on liquidity creation as suggested by Diamond and
Rajan (2000, 2001) depends crucially on deposit insurance coverage being incomplete. If
insurance were complete, depositors would have no incentive to run on the bank, and a
deposit contract would not mitigate the bank’s holdup problem.

Gorton and Winton (2000) show how a higher capital ratio may reduce liquidity creation
through the crowding out of deposits.7 They argue that deposits are more effective
liquidity hedges for investors than investments in equity capital and higher capital ratios
shift investors’ funds from deposits to bank capital. Since deposits are liquid and bank equity
is illiquid, there is a reduction in overall liquidity for investors when the capital ratio is
higher.8

While the theories underlying the “financial fragility-crowding out” hypothesis apply to banks
of all sizes, we argue that they are most applicable to small banks. An essential feature of the
Diamond and Rajan (2000, 2001) model is the bank’s monitoring of the borrower using its
borrower-specific skills, without which the loan’s value diminishes. As the analyses in Berger
et al. (2005) suggest, small banks engage in more relationship-specific lending involving bank
monitoring than large banks. The Gorton and Winton (2000) model is general equilibrium
and theoretically applies to all banks, but as an empirical matter the assumptions of their
model seem to fit small banks better. In their model, there is a single, unsegmented capital
market, so more bank equity capital im- plies less bank deposits. In reality, however, this is
the case only in the smaller markets in which small banks operate. The larger (especially
national and inter- national) capital markets are often quite segmented, with investors
segmenting themselves based on various characteristics, including investment style (e.g.,
choice of equity versus debt instruments). This implies that an increase in the bank’s demand
for equity capital may simply cause “equity investors” to shift out of other equities, rather
than inducing a shift out of bank deposits. That is, changes in bank capital may induce
general equilibrium effects, but these effects may be manifested in securities other than bank
deposits. Further, bank equity capital in practice accounts for a relatively small fraction of
total equity capital held by investors in the national/international capital markets. Thus, even
if the “crowding out” effect by Gorton and Winton (2000) exists for large banks, it is likely to
be empirically smaller than for small banks. Small banks operate in smaller, less segmented
markets, where there is significant overlap between those who invest in bank equity and those
who invest in bank deposits, and any increase in bank capital is more likely to crowd out deposits.
A final and related reason why the crowding out effect may be stronger for small banks is that
these banks fund themselves largely with deposits and capital. In contrast, large banks also
use other liabilities that are less liquid than deposits (such as subordinated debt), suggesting
that an increase in capital may lead to a drop in other liabilities rather than deposits. Thus,
capital is more likely to crowd out deposits at small banks than at large banks.

Under the alternative “risk absorption” hypothesis, which is directly linked to the risk-
transformation role of banks, higher capital enhances banks’ ability to create liquidity. This
insight is based on two strands of the literature. One strand consists of papers that argue that
liquidity creation exposes banks to risk (e.g., Allen and Santomero 1998; Allen and Gale 2004).
The more liquidity that is created, the greater is the likelihood and severity of losses
associated with having to dispose of illiquid assets to meet the liquidity demands of customers.
The second strand consists of papers that posit that bank capital absorbs risk and expands
banks’ risk-bearing capacity (e.g., Bhattacharya and Thakor 1993; Repullo 2004; Von Thadden
2004; Coval and Thakor 2005). Combining these two strands yields the prediction that higher
 capital ratios may allow banks to create more liquidity.

We argue that the risk absorption hypothesis applies more strongly to large banks than to
small banks. The reasons are threefold. First, large banks are typically exposed to more
intense regulatory scrutiny, increasing the value of capital to the banks as part of its
overall risk management. Second, large banks are also subject to greater market discipline
from uninsured providers of finance, so capital has a greater effect on both the cost and the
availability of uninsured financing. Third, some large banks may see new opportunities to offer
large loan commitments or engage in other off-balance sheet activities. Since these activities
involve risk, these banks may boost equity capital in anticipation of engaging in these new
activities.

Finally, we point out one additional contribution of some of the recent theories. The standard
view of liquidity creation is that banks create liquidity by transforming illiquid assets into
liquid liabilities. Diamond and Rajan (2000, 2001) and Gorton and Winton (2000) show,
however, that banks can create more or less liquidity by simply changing their funding mix
on the liability side. Thakor (1996) shows that capital may also affect banks’ asset portfolio
composition, thereby affecting liquidity creation through a change in the asset mix. Our
measures of liquidity creation incorporate these insights—they explicitly recognize that
liquidity creation by banks occurs through changes in the mixes on both sides of the balance
sheet, as well as through off-balance sheet activities.

1.3 Bank capital and liquidity creation: the existing empirical evidence

Some empirical studies examine issues related to liquidity creation, but do not focus on
the role of capital. Kashyap, Rajan, and Stein (2002) provide empirical evidence of
synergies between commitment lending and deposits, consistent with their model. Gatev,
Schuermann, and Strahan (2006) and Gatev and Strahan (2006) find that banks have a
comparative advantage in hedging liquidity risk in the economy because banks experience
deposit inflows following a market crisis or liquidity shock that allow them to have more
funds to provide the additional loans drawn down under commitments at such times.
Pennacchi (2006) confirms the existence of synergies between loan commitments and
deposit taking, but finds that such synergies do not hold prior to the creation of FDIC deposit
insurance. These studies do not focus on the role of bank capital, but they do in some cases
include the capital ratios in regressions of some liquidity creation components, yielding
ambiguous predictions related to the effect of capital on liquidity creation.9

The credit crunch literature tests hypotheses about bank capital and one type of liquidity
creation, usually business lending or real estate lending, during the early 1990s when bank
lending declined significantly. Several studies find that the decline in bank capital ratios
arising from loan losses in the late 1980s and early 1990s contributed significantly to the
reduction in lending (e.g., Peek and Rosengren 1995). This is consistent with a positive
relationship between capital and liquidity creation during a period of distress. In the early
1990s, U.S. regulators also imposed new leverage requirements, as well as the Basel I
risk-based capital standards. Most of the studies found that the leverage requirements
contributed to the decline in lending, consistent with the hypothesis of a negative effect of bank
capital on liquidity creation (e.g., Berger and Udell 1994; Hancock, Laing, and Wilcox 1995;
Peek and Rosengren 1995), and generally concluded that the risk-based capital requirements had
little effect on lending. Unfortunately, the unusual combination of several major changes in bank
capital regulation and a recession makes it difficult to parse the different effects and draw
general conclusions.

Finally, some studies of bank lending behavior include capital ratios, but focus on other
 issues. For example, Berger and Udell (2004) study procyclical lending and find positive,
statistically significant effects of capital on the annual growth of business loans. Holod and
Peek (2007) examine monetary policy effects and find that the capital ratio has significant
positive effects on loan growth. Gambacorta and Mistrulli (2004) use Italian data and find
that the impact of monetary policy and GDP shocks on bank lending depends on bank
capitalization.

Thus, the existing empirical literature sheds relatively little light on the relationship between
bank capital and liquidity creation. Some studies test the liquidity creation theories, but do
not focus on the role of bank capital. Others include capital in their regressions, but specify
only limited components of liquidity creation, and often under unusual circumstances. Our
empirical analysis uses a significantly different approach.

2. Construction of Our Liquidity Creation Measures

In this section, we pursue our first main goal of developing measures of liquidity creation. We
explain the construction of our four liquidity creation measures and clarify which is our
preferred measure. We also show how Deep and Schaefer’s (2004) liquidity transformation
measure can be viewed as a special case of one of our measures.

We construct the liquidity creation measures using a three-step procedure. In step 1, we

classify all bank balance sheet and off-balance sheet activities as liquid, semiliquid, or
illiquid. In step 2, we assign weights to the activities classified in step 1. In step 3, we
combine the activities as classified in step 1 and as weighted in step 2 in different ways to
construct our four liquidity creation measures: “cat fat,” “mat fat,” “cat nonfat,” and “mat
nonfat.” Table 1 illustrates the three-step procedure. We discuss these steps in turn.

2.1 Step 1: Classifying activities as liquid, semiliquid, or illiquid.

In step 1, we classify all assets as liquid, semiliquid, or illiquid based on the ease, cost, and time
for banks to dispose of their obligations to obtain liquid funds to meet customers’ demands.
We similarly classify bank liabilities plus equity as liquid, semiliquid, or illiquid, based on the
ease, cost, and time for customers to obtain liquid funds from the bank. Off-balance sheet
guarantees and derivatives are classified consistently with treatments of functionally similar
on-balance sheet items.10

Ideally, we would use information on both product category and maturity to classify all
bank activities. For example, as noted above, business loans are generally more illiquid
than residential mortgages and consumer loans, as the latter can often be more easily
securitized and sold to meet liquidity demands. Within each category, shorter maturity
items are more liquid than longer maturity items because they self-liquidate without effort or
cost sooner. For bank activities other than loans, Call Reports provide sufficient detail on
category and maturity, so our classifications incorporate both aspects. Un- fortunately, this is
not the case for loans. Call Reports split loans into various loan categories and into different
maturity classes, but do not provide maturity information for individual loan categories. We
therefore either classify loans entirely by category (“cat”) or entirely by maturity (“mat”).
Thus, our “cat” and “mat” liquidity creation measures constructed below classify loans
either by category or maturity, but in all cases incorporate both key characteristics for other
bank activities.

2.1.1 Assets

Classifying loans
• Category (“cat”): For the “cat” measures of liquidity creation, we classify business loans and
leases as illiquid assets, because these items typically cannot be sold quickly without
incurring a major loss. Residential mortgages and consumer loans are generally relatively
easy to securitize, and loans to depositories and governments are likely to be comparatively
easy to sell or otherwise dispose of because the counterparties are relatively large and
informationally transparent. We classify these loan categories as semiliquid assets.

• Maturity (“mat”): Shorter maturity items are more liquid than longer maturity items because
they self-liquidate sooner. We therefore classify all short-term loans of up to one year as
semiliquid and all long- term loans of over one year as illiquid for the “mat” measures.

Classifying assets other than loans: We classify premises and investments in unconsolidated
subsidiaries as illiquid assets, because typically these items cannot be sold quickly without
incurring a major loss. We classify cash, securities, and other marketable assets that the bank
can use to meet liquidity needs quickly without incurring major losses as liquid assets.
Liabilities plus equity

• Classifying liabilities: We count funds that can be quickly withdrawn with- out penalty by
customers, such as transactions deposits, savings deposits, and overnight federal funds
purchased, as liquid liabilities. We classify deposits that can be withdrawn with slightly
more difficulty or penalty as semiliquid. This includes all time deposits regardless of
maturity. We do not differentiate between short-term and long-term time deposits since all
time deposits can be borrowed against with a penalty regardless of maturity. We also
classify as semiliquid the balance sheet item “other borrowed money,” which contains other
short and medium maturities with terms longer than overnight, such as term federal funds,
repurchase agreements, and borrowings from Federal Reserve Banks and Federal Home
Loan Banks. We classify long-term liabilities that generally cannot be withdrawn easily or
quickly, such as subordinated debt, as illiquid.

• Classifying equity: We classify equity as illiquid because investors cannot demand liquid
funds from the bank and the maturity is very long. Although the equity of some banks is
publicly traded and may be sold relatively easily, the investors are able to retrieve liquid
funds through the capital market, not from the bank. Thus, while traded equity may be liquid
from an individual investor’s point of view, such liquidity is created by the capital market,
rather than by the bank, the focus of this article.

2.1.2 Off-balance sheet activities

• Classifying guarantees: We classify loan commitments and letters of credit as illiquid

guarantees. These items are functionally similar to on-balance sheet business loans in that
they are obligations that are illiquid from the point of view of the bank; except in very
unusual circumstances, the bank must provide the funds to the customer upon demand.12 As
well, in most cases, the bank cannot sell or participate these items. We classify net credit
derivatives (i.e., the amount guaranteed minus the beneficiary amount) and net securities lent
(i.e., the amount lent minus the amount borrowed) as semiliquid guarantees since they can
potentially be sold or participated, analogous to semiliquid on-balance sheet residential
mortgages and consumer loans. We classify net participations acquired from other
institutions (i.e., the amount acquired minus the amount conveyed to others) as liquid
guarantees, since they are functionally similar to on-balance sheet liquid securities.

• Classifying derivatives: We classify all derivatives (other than credit derivatives, which we
classify above as guarantees)—interest rate, foreign exchange, and equity and commodity
derivatives—as liquid because they can be bought and sold easily and are functionally
 similar to liquid securities. We focus on the gross fair values of these derivatives (which are
sometimes positive and sometimes negative), which measure how much liquidity the bank is
providing to or absorbing from the public.

2.2 Step 2: Assigning weights to the activities classified in step 1.

In step 2, we assign weights to all of the bank activities classified in step 1. That is, we assign
weights to the classes of liquid, semiliquid, and illiquid assets, liabilities plus equity, and off-
balance sheet guarantees and derivatives shownin Table 1.

We base the weights on liquidity creation theory. According to this theory, banks create
liquidity on the balance sheet when they transform illiquid assets into liquid liabilities. An
intuition for this is that banks create liquidity because they hold illiquid items in place of the
nonbank public and give the public liquid items. We therefore apply positive weights to both
illiquid assets and liquid liabilities, so when liquid liabilities (such as transactions deposits)
are used to finance illiquid assets (such as business loans), liquidity is created. Following
similar logic, we apply negative weights to liquid assets, illiquid liabilities, and equity, so that
when illiquid liabilities or equity is used to finance a dollar of liquid assets (such as treasury
securities), liquidity is destroyed. Note that the negative weight on equity only captures the
direct effect of capital on liquidity creation. Any indirect (positive or negative) effects on
liquidity creation are attributed to the individual items that are affected. For example, if
capital allows banks to extend more illiquid loans, this positive effect is captured by the
positive weight applied to illiquid loans multiplied by the associated dollar increase in loans.

The magnitudes of the weights are based on simple dollar-for-dollar adding- up constraints,
so that $1 of liquidity is created when banks transform $1 of illiquid assets into $1 of
liquid liabilities. Similarly, we require that $1 of liquidity is destroyed when banks transform
$1 of liquid assets into $1 of illiquid liabilities. Based on these constraints, we assign a weight of
½ to both illiquid assets and liquid liabilities and a weight of ½ to both liquid assets and
illiquid liabilities. Thus, when a dollar
− of liquid liabilities (such as transactions deposits) is
used to finance a dollar of illiquid assets (such as business loans), liquidity creation equals ½
∗ $1 ½ ∗ $1 $1. In this case, maximum liquidity ($1) is created. Intuitively, the weight of ½

applies to both illiquid + assets and = liquid liabilities, since the amount of liquidity created is
only “half” determined by the source or use of the funds alone—both are needed to create
liquidity. Similarly, when a dollar of illiquid liabilities or equity is used to finance a dollar of
liquid assets (such as treasury securities), liquidity creation equals ½ ∗ $1 ½ ∗ $1 $1, as
maximum liquidity is destroyed.

Using these weights, banks do not create liquidity when they use liquid liabilities (e.g.,
transaction deposits) to finance liquid assets (e.g., treasuries), or when they use illiquid
liabilities or equity to finance illiquid assets (e.g., business loans). In these cases, banks hold
items of approximately the same liquidity as they give to the nonbank public.

We apply the intermediate weight of 0 to semiliquid assets and liabilities, based on the
assumption that semiliquid activities fall halfway between liquid and illiquid activities. Thus,
the use of time deposits to fund residential mort- gages would yield approximately zero net
liquidity creation, since the ease, cost, and time with which the time depositors may access
their funds early and demand liquidity roughly equals the ease, cost, and time with which the
bank can securitize and sell the mortgage to provide the funds.

We apply weights to off-balance sheet guarantees and derivatives using the same principles,
consistent with the functional similarities to on-balance sheet items discussed in step 1. For
example, illiquid off-balance sheet guarantees (such as loan commitments) are functionally
similar to on-balance sheet illiquid loans (such as business loans) in that they are obligations of
the bank to provide funds that cannot be easily sold or participated. We therefore apply the
 same weight of ½ to illiquid guarantees as we do to illiquid assets. Similarly, we apply the
same weight of 0 to semiliquid guarantees as we do to functionally similar semiliquid on-
balance sheet assets, and we apply the same weight of -½ to liquid guarantees that we do to
functionally similar on-balance sheet liquid assets.

Analogously, the gross fair values of derivatives are assigned the same weight of −½ as on-
balance sheet liquid assets.13 As discussed in step 1, these contracts can be bought and
sold easily and are functionally similar to liquid securities. Like securities, derivatives with
gross positive fair values reduce bank liquidity creation as the bank effectively holds a
valuable liquid asset in place of the public. Derivatives with gross negative fair values
increase bank liquidity creation as the bank effectively holds a negatively valued liquid asset
in place of the public. Since the Call Reports assign positive values to contracts with gross
positive fair values and negative values to those with gross negative fair values, we capture
these opposing effects on liquidity creation by simply applying weights of ½ to the dollar
values of−both.

We arrange the columns in Table 1 such that all the bank activities that contribute to
liquidity creation are on the left, all those that subtract from liquidity creation are on the
right, and all those with an approximately neutral effect on liquidity creation are in the
center. Thus, those that are assigned a weight of ½ —illiquid assets, liquid liabilities, and
illiquid guarantees—are grouped together on the left. Liquid assets, illiquid liabilities plus
equity, and liquid guarantees and derivatives, which are assigned a weight of 1/2, are grouped
− guarantees with 0 weights are grouped
on the right. Finally, semiliquid assets, liabilities, and
in the center.

2.3 Step 3: Constructing liquidity creation measures by combining activities as classified in

step 1 and as weighted in step 2.

In step 3, we combine the activities as classified and weighted in step 1 and step 2, respectively,
in different ways to construct our liquidity creation measures. The measures are similar in
that they all classify activities other than loans using information on product category and
maturity, as discussed in step 1. The measures differ in that we alternatively classify loans
by category or maturity (“cat” versus “mat”), and—to gauge how much liquidity banks create
off the balance sheet—alternatively include or exclude off-balance sheet activities (“fat”
versus “nonfat”). Hence, we have four measures: “cat fat,” “cat nonfat,” “mat fat,” and “mat
nonfat.” The formulas are shown in Table 1. In Table 1, we again arrange the bank activities
that add to liquidity creation on the left, those that subtract from liquidity creation on the
right, and those with an approximately neutral effect in the center. For all measures, we
multiply the weights of 1/2, 1/2, or 0, respectively, times the dollar amounts of the
corresponding bank activities and add the weighted dollar amounts to arrive at the total dollar
value of liquidity creation at a particular bank. We sum across all banks to obtain the total dollar
value of liquidity created by the entire industry.

We recognize that our liquidity creation measures are rough approximations. We classify all
bank activities as liquid, semiliquid, or illiquid, and use three weights, 1/2, 0, and 1/2.
Differences in liquidity obviously exist within each of the three classifications, but the data
generally do not allow for much finer distinctions, and there are no other unambiguous
weights to apply. The use of 1/2, 1/2, and 0 are the clear demarcations of full liquidity, full
illiquidity, and neutrality, respectively, and no other clear choices present themselves.

Note that Deep and Schaefer’s (2004) LT gap measure is conceptually close to our “mat
nonfat” measure and may be viewed as a special case of it. If we classified all assets and
liabilities as either liquid or illiquid (none as semiliquid) using maturities, excluded off-balance
sheet activities, and specified assets (A) rather than gross total assets (GTA), our “mat nonfat”
formula reduces to their formula.
 We next discuss why we consider “cat fat” to be our preferred liquidity creation measure.
First, we argue that the “cat” measures are preferred to the “mat” measures primarily
because what matters to liquidity creation on the asset side is the ease, cost, and time for
banks to dispose of their obligations to obtain liquid funds. The ability to securitize loans is
closer to this concept than the time until self-liquidation. For example, a 30-year residential
mortgage may be securitized relatively quickly even though it is a long-term loan. Second, we
argue that the “fat” measures are preferred to the “nonfat” measures because off-balance
sheet activities provide liquidity in functionally similar ways to on-balance sheet items.
Hence, “cat fat” is our preferred measure.

3. Bank Liquidity Creation over Time, in the Cross Section, and Value Implications

In this section, we pursue our second main goal of gaining a deeper insight into banks’ role
as liquidity creators by applying our four measures to data on the U.S. banking sector. We
first describe how we construct our sample. We then measure how much liquidity banks
create. We next explore the time- series and cross-sectional variation in bank liquidity
creation and examine the characteristics of banks that create the most and least liquidity over
the sample period. In all of these analyses, we split the sample by size. In addition, we divide
the data by bank holding company status, wholesale versus retail orientation, and merger
status. Finally, we explore the value implications of bank liquidity creation.

3.1 Sample description

Our sample includes almost all commercial banks in the United States that are in business
during the 1993 to 2003 period. To ensure that our sample only contains “true,” viable
commercial banks, we impose the following restrictions. We exclude a bank if it (1) has no
commercial real estate or commercial and industrial loans outstanding; (2) has zero deposits;
(3) has zero or negative equity capital in the current or lagged year; (4) is very small
(average lagged GTA below $25 million);17 (5) has unused commitments exceeding four
times GTA; (6) resembles a thrift (residential real estate loans exceeding 50% of GTA); or
(7) is classified by the Federal Reserve as a credit card bank or has consumer loans
exceeding 50% of GTA.18 We also eliminate 0.7% of all bank-year observations because
some of the exogenous variables used in our regression analysis are missing.

For all the banks in our sample, we obtain annual Call Report data as of 31 December of
each year. In all of our analyses, we split the sample by size for several reasons. First, there are
many empirical studies that show that size matters when studying components of bank liquidity
creation. For example, Berger et al. (2005) argue that large and small banks have comparative
advantages in handling different types of credit information, and hence will extend different
types of loans. They split their sample by bank size, and indeed find that large and small
banks make very different loans. Furthermore, Kashyap, Rajan, and Stein (2002) provide
empirical evidence that the relationship between commitments and transactions deposits is
different for banks in different size classes. Second, although there are no theories that argue
that the effect of capital on liquidity creation depends on bank size, we expect that the net
effect of capital on liquidity creation would be different for banks of different size classes.
As shown in Sections 5 and 6, we find confirming empirical evidence. Thus, we split the
sample into large banks (GTA exceeding $3 billion), medium banks (GTA $1 billion–$3
billion), and small banks (GTA up to $1 billion).

Our definition of small banks with GTA of up to $1 billion conforms to the usual notion of
“community banks” that primarily create liquidity by trans- forming locally generated
deposits into local loans on the balance sheet.20 We divide the remaining observations
roughly in half with the $3 billion cutoff for GTA. Large banks with GTA over $3 billion
 create much more liquidity off the balance sheet than small banks. Large institutions also tend
to generate and disperse on-balance sheet funds on more national and international bases than
small institutions. Medium banks with GTA between $1 billion and $3 billion tend to have
portfolios that mix some of the characteristics of small and large banks.

Our sample contains 83,998 bank-year observations: 1804 for large banks, 2132 for medium
banks, and 80,062 for small banks.

3.2 Liquidity creation over time and in the cross section for banks split by size and by bank
holding company status, wholesale versus retail orientation, and merger status.

We next measure how much liquidity banks create and explore how liquidity creation has
changed over time and how it varies in the cross section. We initially split banks only by
size, and then we also divide banks by bank holding company status, wholesale versus
retail orientation, and merger status. Panel A of Table 2 shows the summary statistics on
bank liquidity creation based on our four measures for the entire banking sector and
separately for large, medium, and small banks in 1993 and in 2003, the first and last years of
our sample period, respectively. It also shows graphs of liquidity creation over the entire
sample period using the corresponding measures. As shown, due to consolidation of the
banking industry, the numbers of observations of large and small banks fell by about one-
quarter each, while the number of medium banks remained approximately constant. Overall,
the number of banks in the sample fell by about 23% from 9095 in 1993 to 6968 in 2003.

We find that banks created liquidity of $2.843 trillion in 2003 based on our preferred “cat
fat” measure, which classifies loans by category and includes off-balance sheet activities (see
panel A of Table 2).21 Liquidity creation equals 39% of industry GTA and represents $4.56 of
liquidity per $1 of equity capital. It is about 70% as large as gross loans and 58% as large as
total deposits, which are standard asset and liability measures of bank size. Overall liquidity
creation almost doubled in real terms between 1993 and 2003. As shown, liquidity creation
also increased as a fraction of GTA, equity, gross loans, and total deposits, suggesting that
liquidity creation grew at a faster rate than these items.

Large banks created 81% of industry liquidity at the end of the sample period, despite
representing only about 2% of the sample observations. Medium and small banks generated
only about 5% and 14% of industry liquidity creation as of 2003, respectively, despite their
greater numbers of observations. Medium equity, gross loans, and total deposits than large
banks. As will be shown, this is because these institutions generated much less liquidity off
the balance sheet. At large banks, liquidity creation doubled in real dollars, although it only rose
as a fraction of GTA from 40% in 1993 to 41% in 2003. Perhaps surprisingly, small banks
showed the greatest increase in liquidity creation divided by GTA, equity, gross loans, and
total deposits.

As shown in the “cat fat” graph in panel A of Table 2, liquidity creation based on this
measure increased in every year from 1993 to 2003. This was primarily driven by large
banks—medium and small banks experienced smaller, nonmonotonic increases in liquidity
creation. The increase in overall liquidity creation was driven by substantial growth in
illiquid assets, liquid liabilities, and illiquid guarantees, which outweighed smaller increases
in liquid assets, illiquid liabilities, and equity.

Liquidity creation is almost 50% less based on our “cat nonfat” measure, which is the same as
“cat fat” except for the exclusion of off-balance sheet activities. Large banks still created
most of the industry’s liquidity, although the percentage is lower (71% as of 2003 versus
81% based on our “cat fat” measure). As shown in the “cat nonfat” graph, liquidity
creation based on this measure also increased in every year of the sample period, primarily
due to increases by large banks. Liquidity creation by medium and small banks experienced
smaller, nonmonotonic increases in liquidity creation.

A comparison of liquidity creation based on the “cat fat” and “cat nonfat” measures reveals
= banks create almost half of their liquidity off the balance sheet. This highlights the
that
importance of including off-balance sheet activities. Although banks engage in a variety of
off-balance sheet activities, the main drivers of off-balance sheet liquidity creation are
illiquid guarantees of $2.781 trillion, in particular unused commitments ($2.426 trillion) and,
to a lesser extent, net standby letters of credit ($0.287 trillion).22 Derivatives ($0.023 trillion)
are not among the major components of off-balance sheet liquidity creation. As noted in
Section 2.2, while banks may have substantial derivatives portfolios, most operate with nearly
matched books, so gross positive and negative fair values tend to offset each other. As an
interesting side note, we also find that unused commitments, C&I commitments (a subset of
unused commitments), net standby letters of credit, and commercial and similar letters of
credit are all highly positively and significantly correlated with transactions deposits (ρ 0.73,
0.81, 0.80, and 0.68, respectively), consistent with the predictions and findings of Kashyap,
Rajan, and Stein (2002).

A second insight gained from comparing liquidity creation based on our “cat fat” and “cat
nonfat” measures is that large, medium, and small banks create liquidity in very different
ways. For example, as of 2003, unused loan commitments equal 48% of total liquidity
created by large banks, while only amounting to 26% and 19% of liquidity created by
medium and small banks, respectively. Commercial real estate, on the other hand, equals only
12% of total liquidity creation for large banks, while equaling 36% and 42% of liquidity
creation for medium and small banks, respectively. Similarly, for large banks, transactions
deposits equal only 9% of total liquidity creation, whereas the corresponding figures for
medium and small banks are 15% and 31%, respectively (not shown for reasons of brevity).

We now turn to liquidity creation based on our “mat” measures. Liquidity creation is the
highest in all years using our “mat fat” measure, which differs from our preferred “cat fat”
measure by using loan maturities in place of categories to classify loans. Treating all loans
with maturity of at least one year as illiquid assets increases measured liquidity creation
primarily because most residential mortgages are classified as illiquid (weight 1/2). Recall that
=
these mortgages are classified as semiliquid (weight 0) in the “cat fat” measure because they
=
are relatively easy to securitize. The “mat fat” pattern of liquidity creation over time is similar
to the “cat fat” pattern. The “mat nonfat” measure, which uses loan maturities and excludes
off-balance sheet activities, yields much smaller measured liquidity creation. The “mat
nonfat” liquidity creation pattern resembles the pattern of the other measures, increasing in
all periods, driven by the large banks.

To understand more deeply how liquidity creation has changed over time and how it varies in
the cross section, we split banks in each size class based on three additional characteristics. First,
we divide banks by bank holding company status into multibank holding company (MBHC)
members, one-bank holding company (OBHC) members, and independent banks. Second, we
divide banks by wholesale versus retail orientation, defined here as having below-average
and above-average numbers of branches for their size class, respectively. Third, we split banks
by merger status: those that did and did not engage in mergers and acquisitions during the
prior three years. For each subsample, we show the numbers of banks in 1993 and 2003, and
present graphs that high- light how liquidity creation has changed over time (see panel B of
Table 2). For brevity, we focus on liquidity creation based on our preferred “cat fat”
measure.
 Panel B of Table 2 contains the results. As shown on the left, the vast majority of large and
medium banks are in MBHCs, while small banks are more evenly divided among the three
governance structures. As the graphs make clear, MBHC members created most of the
overall industry liquidity creation, and these banks also experienced the greatest increase in
liquidity creation. Within each size class, MBHC members also created the most liquidity.

As shown on the top right in panel B of Table 2, most of the banks have wholesale
orientation by our definition, but retail banks create most of overall industry liquidity. This
result is driven by large and small banks; among medium banks, liquidity created by retail and
wholesale banks is similar. As shown, liquidity creation grows in each year for both retail
and wholesale banks, except for the spikes in 2002 and 2003. These spikes occur because
Citibank shifted from a wholesale bank to a retail bank in 2002 and back to wholesale status
in 2003.

Finally, as shown in the bottom right in Table 2, panel B, most banks did not engage in M&As,
but most of overall industry liquidity is created by recently merged institutions. This result is
purely driven by large banks; among medium and small banks, institutions that did not engage
in recent merger activity create more liquidity than those that did. This explains why
liquidity creation by recently merged banks increased in almost every year, whereas liquidity
creation remained relatively constant over the sample period for banks that did not engage in
M&As.

3.3 Characteristics of banks that create the most and least liquidity.

We next examine the characteristics of banks that create the most and least liquidity. In each
size class, we split banks into “high liquidity creators” and “low liquidity creators” based on
our preferred “cat fat” measure. We define high and low liquidity creators as those in the top
25% and bottom 25%, respectively, based on (1) overall liquidity creation; (2) liquidity
creation divided by GTA; and (3) liquidity creation divided by equity.

The top, middle, and bottom parts of panel C in Table 2 show the results based on overall
liquidity creation (LC), LC/GTA, and LC/EQ, respectively. Each part shows the average
amount of liquidity created by high and low liquidity creators, and some key characteristics
(BHC status, wholesale versus retail orientation, and merger status) of these banks. For
example, among large banks, each of the high liquidity creators on average created $36
billion in liquidity. Splitting these high liquidity creators by BHC status, we find that 99% of
these banks are MBHC members, and 1% are OBHC members. Similarly, 25% have a
wholesale orientation, while 25% are retail banks; 74% engaged in recent M&A activity,
while 26% were not involved in recent M&A activity. Several findings are noteworthy. First,
not surprisingly, high liquidity creators create substantially more liquidity than low liquidity
creators in each size class. What may be surprising, however, is just how small the
numbers are for the low liquidity creators. In particular, the bottom 25% of small banks in
terms of overall liquidity creation create slightly negative liquidity. This raises the question
of whether these institutions should still be considered to be banks. To address this question,
it is important to recall that banks perform two central roles in the economy, liquidity
creation and risk transformation. While these banks may not create liquidity, they may still
provide valuable risk-transformation services, although a deeper investigation of this issue
is beyond the scope of this article.

Second, MBHC members tend to create the most liquidity in every size class by every
measure of liquidity creation. In all cases, OBHC members and independent banks tend to be
more prevalent among the low liquidity creators.

Third, based on overall liquidity creation, retail banks tend to be high liquidity creators in every
 size class. Maybe surprisingly, we find opposite results when we split banks based on
liquidity creation divided by GTA and equity. One explanation may be that retail banks tend
to be the largest banks in each size class. While these banks create substantial amounts of
liquidity, they create far less liquidity per dollar of assets or equity. Wholesale banks tend to
be low liquidity creators in every size class.

Fourth, a far more diverse picture arises when we look at banks’ M&A history. Among large
banks, high liquidity creators tend to be banks with recent M&A activity, while low liquidity
creators are approximately evenly distributed among those that did and those that did not
engage in M&As. Since most of the small banks did not engage in recent M&As, it is not
surprising that among these banks, most of both the high and low liquidity creators had no recent
M&A activity. However, it is clear that small banks that did engage in M&As in the prior
three years are better represented among the high liquidity creators. The medium bank pattern
falls somewhere in between the patterns for large and small banks.

3.4 Value implications of bank liquidity creation

We next investigate the value implications of bank liquidity creation. If liquidity creation creates
a net surplus to be shared between the bank, its borrowers, and its depositors, then liquidity
creation should be positively associated with the market value of the bank or its holding
company. To examine this issue, we focus on banks that are individually traded or part of a
traded bank holding company. For the purposes of this analysis, we include listed independent
banks and OBHCs, and we aggregate the liquidity creation of all the banks in a listed MBHC.
To ensure that any relationship between liquidity creation and value is likely to be due to the
liquidity created by our sample banks, we exclude holding companies in which these banks
account for less than 90% of holding company assets.23 Imposing this restriction reduces our
sample from 3686 to 3223 bank-year observations.

Since we are not aware of any theories that predict a causal link between liquidity creation
and value, we focus on correlations. In particular, we present correlations between liquidity
creation and value, where liquidity creation is measured by the dollar amount of liquidity
creation and liquidity creation divided by GTA and equity (all calculated using our “cat fat”
measure), and value is measured as the Market-to-Book ratio and the Price-Earnings ratio
(based on earnings before and after extraordinary items).

Table 3 contains the results. As shown, the dollar amount of liquidity creation and liquidity
creation divided by GTA and equity are all significantly positively correlated with the market-
to-book ratio, with correlations between 0.115 and 0.164. The correlations with the price-
earnings ratio (based on earnings before and after extraordinary items) are also all positive,
but they are smaller in magnitude, and are significant in only four of six cases. These results
suggest that banks that create more liquidity are valued more highly by investors.

4. Analytical Framework

We next turn to our third main goal of analyzing the effect of bank capital on liquidity
creation. In this section, we describe our regression framework and explain our regression
variables. In Section 5, we present our empirical results, and in Section 6, we examine the
robustness of our results.

Before we present our regression framework, we note that the theories suggest a causal link
between capital and liquidity creation. According to the “financial fragility-crowding out”
hypothesis, the effect of capital on liquidity creation is negative, and according to the “risk
absorption” hypothesis the effect is positive. The negative effect of capital on liquidity
creation as suggested by Diamond and Rajan (2000, 2001), i.e., the financial fragility effect,
 only arises if deposit insurance coverage is incomplete. Deposit insurance is indeed
incomplete for banks in all three of our size classes over our sample period: most banks fund
themselves partly with uninsured deposits and with overnight federal funds purchased,
another funding source that can run on the bank.

In practice, capital and liquidity creation are to some extent jointly deter- mined. To mitigate
this potential endogeneity problem, our exogenous variables are lagged values, as discussed
above. Nonetheless, this may not be sufficient. We therefore interpret all our regression
results with care. We do not claim to establish causation in our regressions, but at a
minimum, our results yield interesting correlations between capital and liquidity creation
that are consistent with the theories and are robust to a variety of checks, including tests that
involve instrumental variables for capital.

4.1 Regression framework

To examine whether the “financial fragility-crowding out” effect or the “risk ab- sorption” effect
empirically dominates, we use panel datasets on large, medium, and small banks from 1993 to
2003. We regress the dollar amount of bank liquidity creation (calculated using our four
liquidity creation measures) divided by GTA on the lagged equity capital ratio while
controlling for other factors that may affect bank liquidity creation. Normalization by GTA
is necessary to make the dependent variables meaningful and comparable across banks and to
avoid giving undue weight to the largest institutions. Use of dollar amounts of liquidity
creation without normalization would primarily amount to a regression of bank size on capital
and other exogenous variables because banks differ so greatly in size even within each size
class.

Our control variables include bank risk, bank size, BHC membership, merger and acquisition
history, and local market competition and economic environment, as explained in detail
below. We include bank fixed effects to account for average differences over time across banks
that are not captured by the other exogenous variables and to reduce correlations across error
terms. Time fixed effects are added to control for average differences in liquidity creation
across years that are not captured by the other exogenous variables, and to reduce serial
correlation problems. All regressions are estimated with robust standard errors, clustered by
bank, to control for heteroskedasticity, as well as possible correlation among observations of
the same bank in different years.

Table 4 gives descriptions and summary statistics for the exogenous variables. All financial
values are expressed in real 2003 dollars using the implicit GDP price deflator.25 The
exogenous variables are lagged values created using annual data averaged over the three years
prior to observation of the dependent variables to reduce potential endogeneity problems, as
lagged values are more likely to reflect earlier decisions.26 The use of three-year averages, rather
than a single lagged year, also reduces the effects of short-term fluctuations and problems with
the use of accounting data. As well, portfolio changes take time to occur and likely reflect
decisions made on the basis of historical experience, so three years of data may more accurately
reflect the inputs into liquidity creation decisions.27 All of the annual lagged values are merger
adjusted. We collect information from the Federal Reserve Board’s National Information
Center (NIC) database on a bank’s prior M&As, and use it to construct historical pro forma
values.

4.2 Variable descriptions

The key exogenous variable is the lagged capital ratio. For our main analysis, we use EQRAT,
the ratio of equity to GTA. Equity meets the most straightforward, narrow definition of capital
as funds that cannot be easily withdrawn. GTA is the simplest measure of bank size,
although it excludes off-balance sheet activities. GTA equals total assets plus the allowance
for loan and lease losses and the allocated transfer risk reserve—two reserves held for
potential credit losses—so that the full value of the loans financed and liquidity created by the
bank on the asset side are included. In Section 6, we perform two robustness checks on the
capital ratio. First, we replace EQRAT with an alternative capital ratio. Second, realizing that
to some extent, a bank chooses its capital ratio, we use an instrumental variable approach to
resolve any potential endogeneity problems.

We control for bank risk using three risk measures.28 The first measure is earnings volatility
(as in Berger et al. 2008, and Laeven and Levine forthcoming). EARNVOL is measured as
the standard deviation of the bank’s return on assets over the previous twelve (minimum:
eight) quarters. The second measure is a bank’s credit risk, a key risk of banks.
CREDITRISK is calculated as a bank’s Basel I risk-weighted assets and off-balance sheet
activities divided by GTA. The third risk measure is the z-score, which indicates a bank’s
distance from default (e.g., Boyd, Graham, and Hewitt 1993). ZSCORE is measured as a
bank’s return on assets plus the equity capital/GTA ratio divided by the standard deviation of
the return on assets. A higher z-score indicates that a bank is more stable. It is important to
appropriately control for bank risk because a primary reason why banks hold capital is to
absorb risk. The inclusion of risk measures helps to isolate the role of capital in supporting the
liquidity creation function of banks from the role of capital in supporting banks’ function as
risk transformers. It is standard practice in the literature to include risk measures one at a
time.29 Nonetheless, we include all three risk measures in every regression in order to
capture all of the information in all three measures in a single specification. However,
because all three measures assess the same underlying unobservable variable (risk), there is a
serious multicollinearity problem associated with introducing them simultaneously. We deal
with this problem by orthogonalizing two of the risk measures.30 In all of the analyses
reported in the article, we include EARNVOL, orthogonalized CREDITRISK, and
orthogonalized ZSCORE. (For simplicity, we use the terms CREDITRISK and ZSCORE
instead of orthogonalized CREDITRISK and orthogonalized ZSCORE throughout.) Since our
approach is a departure from the standard approach of introducing risk measures one at a
time, we also run the base regressions by alternatively including these three risk measures
individually. The results are qualitatively similar to those reported in the article and are
available upon request.

To control for bank size, we include the natural log of bank size, Ln (GTA), in every
regression, as well as running the regressions separately for large, medium, and small banks.
The natural log is used for all of the continuous exogenous variables that may take on large
values to avoid potential specification distortions, given that the dependent variables are
generally in the [0,1] interval.31

We control for the bank’s bank holding company (BHC) status with two variables: D-
MBHC, a dummy variable that equals one if the bank has been part of a multibank bank
holding company (MBHC) in any of the past three years and zero otherwise, and D-OBHC,
a dummy variable that equals one if the bank has been part of a one-bank holding company
(OBHC) in any of the past three years and zero otherwise. BHCs and other banks in the same
BHC may serve as internal capital markets to provide capital in times of stress. This view is
supported by both regulation and the literature.

We also control for the bank’s merger and acquisition history. The D-BANK- MERGE and D-
DELTA-OWN dummies indicate whether a bank was involvedin a merger or acquisition over
the past three years, where a merger is defined as the combination of bank charters into an
institution with a single set of books, and an acquisition is defined as a case in which the
bank’s top-tier holding company changed with no change in charter status. Controlling for
 mergers and acquisitions is important because banks often substantially alter their lending
behavior following such events.

To construct controls for local market competition and economic environment, we define the
local market as the Metropolitan Statistical Area (MSA) or non-MSA county in which the
offices are located.33 For banks with offices in more than one local market, we use weighted
averages across these markets, using the proportion of the bank’s deposits in each of these
markets as the weights. To control for local market competition, we include HERF, the
Herfindahl index of concentration for the market or markets in which the bank is present. We
base HERF on the market shares of both banks and thrift institutions, given that thrifts
compete vigorously with banks for deposits. We also include SHARE-ML, the local market
share of medium and large institutions, to allow for the possibility that banks of different
sizes may compete differently. It is important to control for local market competition
because various studies have shown that market concentration affects credit availability
(e.g., Petersen and Rajan 1995) and that the loan portfolios of large and small banks are
markedly different (e.g., Berger et al. 2005). Hence, competition likely affects liquidity
creation through both the amount and types of loans a bank ex- tends and the way it funds its
activities.35 To control for local market economic conditions, we include the log of
population Ln (POP), the log of population density Ln (DENSITY), and income growth
INC-GROWTH.

5. Regression Results

In this section, we present our regression results. We first present our main results and find
that the relationship between capital and liquidity creation differs for large, medium, and small
banks. We then investigate why the results differ by size class using the components of
liquidity creation. In all cases, we examine whether the findings are consistent with the
economic intuition discussed earlier. In Section 6, we conduct a number of robustness
checks.

Before proceeding, we note the important distinction between the liquidity creation weight
on capital and the regression coefficient on lagged capital. We assign a weight of ½ to equity
when forming our liquidity creation measures, the dependent variables in the regressions. This
does not imply that when we regress the dollar amount of liquidity creation (normalized by
GTA) on the lagged equity ratio, EQRAT, the coefficient on EQRAT should necessarily be
negative or close to 0.5. Rather, the measured effect depends on bank behavior. For example, if
banks with more lagged equity capital extend significantly more illiquid loans and hold
significantly fewer liquid assets than banks with lower levels of capital, we may find a positive
association between lagged capital and liquidity creation.36

5.1 The net effect of capital on liquidity creation for large, medium, and small banks

Panels A, B, and C of Table 5 contain the regression results for large banks (GTA exceeding
$3 billion), medium banks (GTA $1 billion—$3 billion), and small banks (GTA up to $1
billion), respectively. All of our regressions include the full set of control variables and have
time and bank fixed effects, but the results are similar if we control only for size and include
fixed effects (not shown for brevity’s sake).

The results in panel A of Table 5 suggest that for large banks, the relation- ship between
capital and liquidity creation is positive and significant when liquidity creation includes off-
balance sheet activities, i.e., when we use our “fat” liquidity creation measures (“cat fat” or
“mat fat”). The magnitude of the coefficient on the lagged equity capital ratio in the “cat fat”
regression, 1.146, suggests that large banks with a 1 percentage point higher equity capital
ratio for the prior three years (i.e., an increase in EQRAT of 0.01) create additional liquidity
 of over 1 percentage point of a large bank’s GTA, which appears to be a substantial effect.
Using the “nonfat” measures, capital is not significantly correlated with liquidity creation,
suggesting that off-balance sheet activities constitute an important part of the effect of
capital on liquidity creation for large banks. The EQRAT coefficients in the “cat” and “mat”
specifications are of similar magnitude, suggesting that use of maturities in place of
categories for loans has little impact on the measured net effect of capital.

The results for medium banks in panel B of Table 5 are mixed. For these banks, the
relationship between capital and liquidity creation is positive and not significant for the “fat”
measures and negative and marginally significant for the “nonfat” measures.

The results in panel C of Table 5 suggest that for small banks, the relation- ship between
capital and liquidity creation is negative, in sharp contrast to the positive or mixed
relationship found for large and medium banks. All of the coefficients on the lagged capital
ratio are negative and significant at the 1% level, yielding a fairly clear result that is robust
across the liquidity creation measures. Using our preferred “cat fat” measure, the magnitude
of the coefficient on the lagged equity capital ratio, 0.330, suggests that small banks with a
0.01 higher EQRAT create less liquidity by about a third of a percentage point of their GTA.
As for the large banks, the magnitudes of the net effect of capital on liquidity creation are
similar for the “cat” and “mat” measures for small banks. However, a key difference for
small banks is that the “fat” and “nonfat” magnitudes are also similar. The inclusion of off-
balance sheet activities makes little difference to the net effect of capital on liquidity creation,
reflecting the lesser role of these activities for small institutions.

In sum, we find that for large banks, capital and liquidity creation are positively correlated
when we use measures that include off-balance sheet activities, while this relationship is
insignificant when we exclude those activities. For small banks, capital and liquidity creation
are negatively correlated using all of our measures, while for medium banks, the relationship
is mixed. Thus, the data suggest that, consistent with our economic intuition, the “risk
absorption” hypothesis dominates for large banks when off-balance sheet activities are
included and the “financial fragility-crowding out” hypothesis strongly dominates for small
banks. The two effects are largely offsetting for medium banks. We next investigate what
drives these differences.

5.2 Why is the net effect of capital on liquidity creation different by banksize class?

To understand more deeply why the relationship between capital and liquidity creation differs
by bank size class, we examine the relationship between capital and the individual components
of liquidity creation (e.g., liquid, semiliquid, and illiquid assets). Specifically, we use the
individual components based on our “cat fat” liquidity creation measure normalized by GTA
as dependent variables in our regressions.

Panels A, B, and C of Table 6 show the coefficients on EQRAT from these regressions for
large, medium, and small banks, respectively. All of the control variables from the full
specification are included in these regressions, but are not shown in the interest of brevity.
Importantly, since liquidity creation equals the weighted sum of the individual components
(using the ½, 0, and ½ liquidity creation weights discussed above), the weighted sums of the
EQRAT coefficients on the individual liquidity creation components in Table 6 equal the
coefficient on EQRAT using the “cat fat” measure in Table 5. Therefore, the EQRAT
coefficients in the individual component regressions help us understand which components of
liquidity creation yield the different results for large, medium, and small banks.

The results in panel A of Table 6 suggest that for large banks, lagged capital is positively
 related with liquidity creation on the asset side of the balance sheet, as well as off the
balance sheet. Banks with higher lagged capital ratios have significantly more illiquid assets,
fewer liquid assets, and more illiquid guarantees. These positive effects of capital are
partially offset by the fact that large banks with higher lagged capital ratios have significantly
higher capital ratios in the current period (i.e., the coefficient on EQRAT in the equity/GTA
regression is positive and significant). Thus, the positive relationship between lagged capital
and liquidity creation calculated using our “cat fat” measure in panel A of Table 4 is the
net result of the positive relationship between lagged capital and assets and illiquid
guarantees being larger than the negative relationship with current capital. The insignificant
relationship between lagged capital and liquidity creation calculated using the “cat nonfat”
measure in panel A of Table 4 occurs because the positive relationship between lagged
capital and illiquid guarantees is excluded—the positive relationship between lagged capital
and assets approximately offsets the negative relationship with current capital.

The findings in panel A of Table 6 are also consistent with the economic intuition that the
“risk absorption” effect is relatively strong for large banks. Higher capital allows large
banks to bear significantly more portfolio risk, and the data suggest that they do so. Large
banks with higher capital hold more risky illiquid assets such as commercial loans and risky
illiquid guarantees such as loan commitments, and fewer relatively safe liquid assets such as
treasuries. Panel B of Table 6 suggests very different effects for medium banks. Banks with
higher lagged capital ratios tend to have fewer liquid liabilities. The negative relationship
between lagged capital and liquid liabilities approximately offsets the positive relationship
between lagged capital and current capital, yielding the overall insignificant effect for our
“cat fat” measure.

Panel C of Table 6 reveals that similar to large and medium banks, small banks with higher
lagged capital ratios have significantly higher capital ratios in the current period. However, in
stark contrast to large banks, for small banks, capital is negatively related with liquidity
creation on the asset and liability sides of the balance sheet, and essentially unrelated to
liquidity creation off the balance sheet. Small banks with higher lagged capital ratios have
significantly more liquid assets and fewer liquid liabilities. Thus, the effect of lagged capital is
consistently negative for small banks, as opposed to the positive effect for large banks,
because the negative effect on current capital is augmented by negative effects of capital on
the asset and liability sides, and is not offset by a positive effect off the balance sheet.38

The results in panel C of Table 6 are also consistent with the economic intuition that the
“financial fragility-crowding out” effect is relatively strong for small banks. On the asset
side, lagged capital is not positively related with illiquid assets, but instead is positively
related with liquid assets. This is consistent with the spirit of the financial fragility
arguments put forth in Diamond and Rajan (2000, 2001). Capital reduces the financial fragility
needed to commit to monitoring its borrowers. As a result, banks with higher lagged capital
ratios may invest more in liquid assets, rather than increasing their loans. On the liability side,
lagged capital is negatively related to liquid liabilities, consistent with the “crowding out” of
transactions deposits as in Gorton and Winton (2000).

6. Robustness Issues

In Section 5, we found that, based on our preferred “cat fat” measure of liquidity creation, the
relationship between capital and liquidity creation is positive and significant for large banks
(when off-balance sheet activities are included), insignificant for medium banks, and
negative and significant for small banks. We now examine the robustness of these main findings
to (1) using an alternative method to measuring off-balance sheet bank liquidity creation; (2)
using an alternative way of establishing which assets are securitizable; (3) excluding equity
from the liquidity creation measure; (4) using an alternative capital ratio; (5) splitting the sample
 by bank holding company status, wholesale versus retail orientation, and merger status; and (6)
using an instrumental variable approach. We show that our main results are qualitatively
unchanged.

6.1 Using an alternative method to measuring off-balance sheet liquidity creation

Our liquidity creation measures are based on the ease, cost, and time for customers to obtain
liquid funds from the bank, and the ease, cost, and time for banks to dispose of their
obligations in order to meet these liquidity demands. An alternative would be to use the
probability or frequency with which the bank or customers actually liquefy the items and
obtain liquid funds. We argue that the ability or option to obtain funds when needed or
desired is more important than the actual drawdown frequency. This is also what the theories
suggest: banks create liquidity on the balance sheet because they give depositors a liquid
claim to their funds (i.e., the option to withdraw funds when needed) instead of forcing them
to hold illiquid loans directly (e.g., Diamond and Dybvig 1983). Similarly, banks create
liquidity off the balance sheet through guarantees that allow customers the option to draw
down liquid funds when needed (e.g., Kashyap, Rajan, and Stein 2002).

Despite our reservations, we construct a liquidity creation measure that incorporates the
frequency with which customers obtain liquid funds on off- balance sheet guarantees. Our
alternative liquidity creation measure is identical to our “cat fat” measure, except that we
multiply the dollar amount of illiquid off-balance sheet guarantees by 0.30, the observed
frequency of drawdown as documented in recent research (Sufi 2007).39

Using this alternative “cat fat” measure, we find that liquidity creation of the banking sector
is about one-third lower than using our preferred “cat fat” measure in every year and
amounts to $1.869 trillion in 2003 (not shown for brevity). The overall pattern of liquidity
creation, however, is fairly similar to the “cat fat” pattern.

In panel A of Table 7, we regress the dollar amount of liquidity creation using this
alternative “cat fat” measure normalized by GTA on EQRAT and the other exogenous
variables. As shown, based on this alternative method to measuring liquidity creation, we
obtain consistent results: the coefficient on EQRAT is positive for large banks, negative for
small banks, and insignificantfor medium banks.

In principle, this methodology could be applied to all bank activities. For example, the
drawdown frequency is 1 for loans since customers have already received liquid funds.
However, constructing measures using this methodology is difficult, since data on the
frequency of drawdown or sale are unavailable for many activities. More importantly, the use
of drawdown rates goes directly against the liquidity creation theories, which argue that
banks create liquidity by giving customers the option to obtain liquid funds when needed or
desired.

6.2 Using an alternative way of establishing which assets are securitizable.

The amount of liquidity a bank creates is affected by the bank’s ability to securitize its
assets. Our “cat” liquidity creation measures incorporate this by classifying loan categories
that are relatively easy to securitize (residential real estate loans and consumer loans) as
semiliquid and all other loan categories as illiquid. Our “cat” measures do not incorporate,
however, the fact that the ability to securitize assets has developed greatly over our sample
period. In every loan category, a larger fraction of loans was securitized in 2003 than in
1993. We now construct an alternative “cat fat” liquidity creation measure that takes this
development into account.
 Our alternative “cat fat” measure is identical to the “cat fat” measure de- scribed in Section
2, except for the way we classify loans. For each loan category, we obtain year-end U.S.
Flow of Funds data on the total amount of loans outstanding and the total amount of loans
securitized. We use these data to calculate the fraction of loans that has been securitized in the
market in each year. Following Loutskina (2006), we then assume that each bank can
securitize that fraction of its own loans. To give an example, in 1993, $3.1 trillion in
residential and real estate loans were outstanding in the market, and 48.4% of these loans
were securitized. If a bank has $10 million in residential and real estate loans in that year, we
assume that 48.4% thereof can be securitized, and hence, we classify $4.84 millions of these
loans as semiliquid and the remainder as illiquid.

We raise two reservations regarding this alternative approach for our purposes. First, it uses
the actual amount of securitization, whereas the theories suggest that the ability to securitize
matters for liquidity creation, not the amount securitized. Second, this alternative approach
assumes that each bank securitizes the same fraction of loans in a particular category, even
though in practice major differences may exist across banks. That is, when we assume that
48.4% of all residential and real estate loans can be securitized in 1993, one bank may have
securitized virtually its entire residential real estate portfolio in that year, while another bank
may have securitized nothing.

Using this alternative approach, the banking sector created more liquidity, but the growth
pattern is similar to the “cat fat” pattern described in Section 3. Based on this alternative
measure, liquidity creation equaled $1.843 trillion in 1993 and increased by about 70% to
$3.168 trillion in 2003.

The regression results presented in panel B of Table 7 reinforce our prior findings. That is,
for large banks, the relationship between capital and liquidity creation is positive and
significant. For small banks, the relationship between capital and liquidity creation is
negative and significant. For medium banks, the relationship is again statistically
insignificant.

6.3 Excluding equity from the measurement of liquidity creation

Our regression specification is inspired by the theories of bank liquidity creation. These
theories argue that banks create liquidity when illiquid assets are transformed into liquid
liabilities, not when they are transformed into illiquid claims such as equity. The theories also
suggest that equity may affect a bank’s ability to create liquidity. For example, having more
equity capital may allow a bank to extend more illiquid loans. However, as noted in Section
5, a potential concern about our regression specification is that current bank equity is included
(with a weight of ½) in our dependent variables, while the lagged equity ratio is our key
exogenous variable. To ameliorate this potential concern, we create an alternative “cat fat”
liquidity creation measure that excludes equity. This measure does not penalize banks for
funding part of their activities with equity capital. As a result, the measured amount of
liquidity creation is higher for all banks, and this increase is greatest for banks that hold the
most capital. We rerun our regressions using this alternative measure.

The results shown in panel C of Table 7 suggest that our main findings are robust to the
exclusion of equity from our dependent variable. The coefficient on EQRAT is again positive
and significant for large banks, insignificant for medium banks, and negative and significant
for small banks.

6.4 Using an alternative capital ratio

In our main analysis, we use EQRAT, the ratio of equity to GTA, as our key exogenous
 variable. We now replace EQRAT with TOTRAT, the ratio of total capital (as defined in the
Basel I capital standards) to GTA. Total capital includes equity plus limited amounts of other
financial instruments, such as long-term subordinated debt.

One motivation for using this alternative capital ratio is to see if there is a different effect
of regulatory capital from conventional equity capital on liquidity creation. A second
motivation is to allow for a broader definition of capital in line with some of the theoretical
studies. For example, Diamond and Rajan (2000, 2001) indicate that capital in their analysis
may be interpreted as either equity or long-term debt, sources of funds that cannot run on the
bank.

The results based on this alternative capital ratio are shown in Table 8 and are qualitatively
similar to our main results. The relationship between capital and liquidity creation is
positive and significant for large banks, statistically insignificant for medium banks, and
negative and significant for small banks.

6.5 Splitting the sample by bank holding company status, wholesale versus retail orientation,
and merger status

In all of the regression results presented thus far, we have split our sample only by size. In
Section 3, however, we also split our sample by bank holding company status, wholesale
versus retail orientation, and merger status, and showed that substantial time-series and
cross-sectional variation exists among these banks in terms of their ability to create liquidity.
We now test the robust- ness of our main results by rerunning our regressions by size class for
MBHC members, OBHC members, and independent banks; banks with wholesale and retail
orientations; and banks with and without recent M&A activity.

The results are shown in Table 9. For large banks, the coefficient on EQRAT is positive and
statistically significant (except for the small subsample of in- dependent banks, which has
only fifty-four observations). For medium banks, the coefficient on EQRAT is positive and
significant for OBHC members, and not significant for any of the other subsamples. For small
banks, the coefficient is negative in all cases, and significant in all but one case. Thus, our
main findings are generally robust to splitting the data by bank holding company status,
wholesale versus retail orientation, and merger status.

6.6 Using an instrumental variable approach

All the analyses presented so far suggest that, based on our preferred “cat fat” liquidity
creation measure, the relationship between capital and liquidity is positive and significant for
large banks, and negative and significant for small banks. We have been careful thus far to
interpret our results merely as showing strong correlations consistent with the theories rather
than as evidence of a causal relationship, even though our main analysis used three-year
lagged average values of capital to mitigate potential endogeneity concerns.43 We now
address this endogeneity issue more directly. Specifically, we use an instrumental variable
approach to examine whether our statistically significant results about the positive relationship
between capital and liquidity creation for large banks and the negative relationship between
capital and liquidity creation for small banks represent the causal effects of capital on
liquidity creation for large and small banks.

Since we have panel data and use time and bank fixed effects in all of the regressions
presented so far, we should also use fixed effects in our instrumental variable regressions.45 We
can do so if the instruments show sufficient variation over time. Provided this is the case, in
the first stage, we will regress our potentially endogenous variable, EQRAT, on an instrument
and all of the control variables and time and bank fixed effects. In the second stage, we will
regress liquidity creation (using our preferred “cat fat” measure) divided by GTA on the
predicted value for EQRAT from the first stage and all the control variablesand fixed effects.

An instrument must satisfy two requirements. First, to be valid, the instrument should be
correlated with the amount of lagged capital (once the effects of the other exogenous
variables have been netted out) but should not directly affect the amount of liquidity a bank
creates. Second, as noted above, since we want to include bank fixed effects in the
regressions, it is important that the instrument shows sufficient variation within a bank’s
observations over time. We select two instruments: the first one meets both requirements,
while the second one meets only the first requirement. Regressions that include the
instrument that meets both requirements are therefore carried out following the approach
highlighted above. After discussing our instruments, we explain our approach for regressions
that use the other instrument.

Our first instrument is EFF-TAX, the state income tax rate a bank has to pay. Since
interest on debt is tax deductible while dividend payments are not, banks that operate in
states with higher income tax rates are expected to have lower equity ratios, keeping all else
equal. Furthermore, there is no reason to believe that the state income tax rate directly affects
liquidity creation. Similar to Ashcraft (2008), we use the effective income tax rate to be paid on
$1 million in pretax income as our instrument.46,47 If a bank operates in multiple states, we
use the bank’s weighted average state income tax rate, calculated using the share of deposits in
each state (relative to the bank’s total deposits) as weights.48 Our second instrument is
SENIORS, the fraction of seniors (people aged sixty-five and over) in the markets in which a
bank is active. Seniors own larger equity portfolios than the average family. According to the
Survey of Consumer Finances, families headed by a senior were slightly less likely to own
stock (36.1% owned stock versus 47.5% of all families), but the dollar value of the stock
portfolio of those that did own stock was roughly three times as large ($81,200 versus
$27,000 for the average family) (see Bucks, Kennickell, and Moore 2006). Furthermore,
using U.S. data, Coval and Moskowitz (1999) document that investors have a strong
preference for investing close to home. They find that this preference is greater for firms that
are smaller, more highly levered, and those that produce goods that are not traded
internationally. In combination, this evidence suggests that banks—particularly small banks—
that operate in markets with more seniors have easier access to equity financing and hence, will
use more equity financing. We calculate the fraction of seniors using county- and MSA-level
population data from the 2000 decennial Census.

Panel A of Table 10 examines the extent to which the instruments vary. The data suggest that
there is substantial time-series variation for the tax rate for both size classes. We therefore use
the entire sample and include time and bank fixed effects in all the analyses in which we
employ the effective tax rate. The fraction of seniors shows sufficient variation in the cross
section, but not over time since we only have information on the fraction of seniors from the
2000 Census. In the analyses that use SENIORS, we therefore do not use the entire sample
and do not include time and bank fixed effects. Rather, we use liquidity creation data only
from the year 2001 and lagged values of the instrument (i.e., the 2000 Census values) and the
other exogenous variables.

Panel B of Table 10 shows the results of our first-stage regressions. The state income tax
rate has a significantly negative effect on capital for large banks, consistent with Ashcraft
(2008). The tax rate does not significantly affect capital at small banks, potentially because
the tax benefit of debt may be outweighed by safety and soundness considerations that induce
those banks to hold higher capital ratios. The fraction of seniors has a significantly positive
effect on capital for small banks. It does not significantly affect capital at large banks,
possibly because these banks are not limited by geography in terms of their access to a
variety of funding sources.
 Panel C of Table 10 contains the second-stage instrumental variable regression results. For
completeness, we show results for both size classes using both instruments. However, since
the first-stage results showed that the tax rate can only be used as an instrument for large
banks while the fraction of seniors can only be used as an instrument for small banks, we
focus our attention on those two regressions. When we use instruments for capital, our results
are consistent with our earlier findings. The effect of capital on liquidity creation is positive
and statistically significant for large banks, and negative and significant for small banks.

For both large and small banks, the coefficients on EQRAT are larger when we use
instruments, suggesting that the effect of capital on liquidity creation is several times the
previously estimated effect. Using similar logic as in Levitt (1996), this suggests that in our
main liquidity creation regressions, EQRAT is correlated with the residuals, inducing a bias
toward zero in our coefficient estimates. When we use instruments for capital, we obtain
consistent estimates.51 However, since we had no a priori reason to believe that our EQRAT
coefficients were understated in our main regressions, we are hesitant to put too much
weight on this explanation.

7. Conclusion

According to banking theory, banks exist because they create liquidity and transform risk.
Our understanding of the liquidity creation role is hampered by the absence of comprehensive
liquidity creation measures. The first contribution of this article is the development of four
bank liquidity creation measures. Our second contribution is that we use our measures to gain a
deeper insight into banks’ role as liquidity creators. We determine the magnitude of bank
liquidity creation, its intertemporal patterns, its cross-sectional variation, characteristics of
high and low liquidity creators, and examine the relationship between liquidity creation and bank
value. Our third contribution is that we use our measures to study an issue of significant
research and policy relevance—the effect of bank capital on liquidity creation—and thereby
test the predictions of recent theories about the relationship between capital and liquidity
creation.

Our calculations suggest that liquidity creation by the U.S. banking sector exceeded $2.8
trillion as of 2003 based on our preferred liquidity creation measure, and nearly doubled in
real terms between 1993 and 2003. Interestingly, banks create only about half of their liquidity on
the balance sheet, highlighting the importance of off-balance sheet liquidity creation. Large
banks (gross total assets exceeding $3 billion) create 81% of the liquidity while comprising
only 2% of all banks. Multibank holding company members, retail banks, and recently
merged banks create most of the industry’s overall liquidity and show the greatest growth in
liquidity creation over time. Liquidity creation is also positively associated with bank value.

When we test the relationship between capital and liquidity creation, we find empirical support
for both the theories which predict that higher capital may suppress liquidity creation and
those which suggest that higher capital may enhance banks’ ability to create more liquidity.
Based on our preferred “cat fat” liquidity creation measure, the relationship between capital
and liquidity creation is positive and significant for large banks, insignificant for medium
banks, and negative and significant for small banks. We perform a variety of robustness
checks and find consistent results.

Our finding that the relationship between bank capital and bank liquidity creation differs by
bank size raises interesting policy issues. It is well known that regulators impose capital
requirements on banks for safety and soundness reasons. Our findings suggest that while
regulators may be able to make banks safer by imposing higher capital requirements, this
benefit may have associated with it reduced liquidity creation by small banks, but enhanced
liquidity creation by large banks.

Our liquidity creation measures may also be used to address a number of other interesting
issues that are beyond the scope of this article, but may be pursued in future research. Does
liquidity creation affect economic growth? How do monetary policy initiatives by central
banks, changes in deposit insurance, and other policy innovations affect liquidity creation?
How does liquidity creation differ across nations? How much liquidity do banks create compared
to nonbank financial intermediaries? How much liquidity do banks create relative to financial
markets, and what are the complementarities, if any, in liquidity creation between banks and
capital markets? Addressing these questions holds the promise of substantially improving our
understanding of the liquidity creation function of banks and how it affects the economy.