Heterogenous Banks and Macroprudential
Regulations
by
Subhendu Bhowal
B. Tech, M.Tech, IIT Kharagpur (2008)
P.G.P.M, IIM Ahmedabad (2010)
Submitted to the Sloan School of Management
in partial fulfillment of the requirements for the degree of
MASTER OF SCIENCE IN MANAGEMENT RESEARCH
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
February 2019
@ Massachusetts Institute of Technology 2019. All rights reserved.
. Signature redacted
A uthor ........... I................
Sloan School of Management
Signature redacted February 11, 2019
C ertified by ... .........................
Adrien Verdelhan
w
Associate Professor
Thesis Supervisor
AcceAptceed pbye.d..b y..Signature 
redacted ....................
Ezra Zuckerman Sivan
Alvin J. Siteman (1948) Professor of Entrepreneurship and Strategy
Deputy Dean
Professor, Technological Innovation, Entrepreneurship, and Strategic
MASS TECHNL Y Management and Work and Organization Studies
APR 03Z019
LIBRARIES

3
Heterogenous Banks and Macroprudential Regulations
by
Subhendu Bhowal
Submitted to the Sloan School of Management
on February 11, 2019, in partial fulfillment of the
requirements for the degree of
MASTER OF SCIENCE IN MANAGEMENT RESEARCH
Abstract
This paper studies how financial intermediation varies across banks. Bank size is a
first-order determinant of banks' capital structure in the cross-section. Largest banks
have the lowest capital-to-asset ratio and the lowest ratio of Tier-1 capital against
risk-weighted assets. These large banks earn a larger interest income per dollar
invested in their loan portfolio than small banks, and they maintain the highest
net interest margins among all banks. A cash flow sensitivity analysis shows that
the largest banks are the most tightly constrained by minimum capital requirement,
while all other banks maintain capital in excess of minimum capital requirement
regulation. Empirically, banks do not adjust their lending portfolio dollar for dollar as
their net profits increase or lever up immediately by issuing more deposits. Further,
we find that the financial accelerator amplifies productivity shock in aggregate data.
The impulse response to total productivity shock shows that the loan volume of the
capital-constrained largest banks does not respond positively to positive productivity
shocks. This is in contrast to smaller banks that increase loans when productivity
improves in the economy.
Thesis Supervisor: Adrien Verdelhan
Title: Associate Professor

5
Acknowledgments
Firstly, I would like to express my gratitude to Adrien Verdelhan for his continuous
support during the toughest times. I am indebted to him for his kindness, patience
and generosity.
I would like to express my deepest appreciation to Jason McKnight, Assistant
Dean for Graduate Education, who provided me the possibility to complete this
report. I'd like to thank Larry Anderson and Dawn Anderson, and Michael Collins
for making my life easier in Tang Hall of Residence.
I have a great fortune to receive unwavering support from family and friends
that made the journey easier. Sourav Sarkar and Giulio Alighieri have been dear
friends during my time at MIT. I cherish the countless hours I have spent with these
two people. I had several interesting discussions with Fernando Martins, Maarten
Meeuwis, Leonardo Elias and Fangzhou Lu over the years.
I'm especially grateful to my mother Sipra Bhowal for everything in my life.
Deepest gratitude to my girlfriend Srabonti Chakraborty, for all her love and support.
Finally, nothing would have been possible without my father Late Sukhendu
Bikash Bhowal. I miss you, dad!
6
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 D ata source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4 Bank Characteristics and Summary Statistics . . . . . . . . . . . . . 19
5 Investment Cash Flow Sensitivity of Banks . . . . . . . . . . . . . . . 25
6 Estimating Impulse Response in the Data . . . . . . . . . . . . . . . 32
6.1 IRF in Aggregate Time Series Data . . . . . . . . . . . . . . . 33
6.2 IRF in the Panel Data for US Banking . . . . . . . . . . . . . 40
7 C onclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
7
8
List of Figures
1 Average Equity to Asset Ratio by Size Decile . . . . . . . . . . . . . . 21
2 Average Leverage, Interest Income, Interest Expense and NIM by Size
D ecile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3 Average Loan to Asset, Tier-1 Capital to Risk Weighted Asset, Av-
erage Deposit to Asset and Average Non-Deposit Liabilities to A byt
Size D ecile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4 Effects of TFP shock, Jorda local projection, on real variables . . . . 37
5 Effects of TFP shock, Jorda local projection, on price cost margin . . 38
6 Effects of TFP shock, Jorda local projection, on total loans, interest
income, interest expense and average NIM . . . . . . . . . . . . . . . 42
9
10
List of Tables
1 Summary Statistics for All Banks in the entire Panel Data (1983Q1
to 2013Q 4) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2 Summary Statistics for Banks in Highest Decile and Highest Percentile
in the entire Panel Data (1983Q1 to 2013Q4) . . . . . . . . . . . . . . 27
3 Cash Flow Sensitivity of Bank Loans and Deposits Controlling for
10-Year Bond Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4 Cash Flow Sensitivity of Bank Loans and Deposits Controlling for NIM 35
11
1 Introduction
The size of financial intermediaries affects the capital structure decisions. The fund-
ing cost and the average interest income vary systemically across size deciles. Banks
in the highest size deciles are considered systemically important and often considered
too-big-to-fail. Since financial crisis, the systemically important banks have received
both more stringent regulatory scrutiny and inordinate amount of Government sup-
port to avoid failure. As macroeconomic modeling incorporate financial frictions, this
differential effects of size has been largely ignored by equilibrium models with finan-
cial frictions. This raises the question: How does size affect financial intermediation
in data?
The objective of this paper is to examine whether there bank size is a first order
determinant of banks capital structure in the cross-section. I examine whether banks
are truly constrained by capital requirements. I further investigate how productivity
shock propagates differently across different sizes of bank.
I find that several bank capital structure variables show clear relationship with
sizes. Average equity to asset is declining in size and alternatively, average book
leverage ratio is increasing in size. The average interest income and average net
interest margin for the highest decile is sharply higher than the other banks. Also
average loan to asset ratio, average non-deposit liabilities increase with size. Tier-1
capital to risk weighted asset falls with size.
I estimate cash flow sensitivity for the banks along the strategy Fazzari et al.
(1988). I show that the regression coefficients of contemporaneous loan and next
quarter loan net profit is increasing in size and lines up perfectly. I find that the same
monotonicity in the coefficients for deposit on net profit is not obtained. The results
show that banks are constrained by capital, however, the tightness increases with
12
size. Also, as bank net worth improves, the banks do not adjust lending immediately
dollar for dollar, or cannot increase deposit taking to the tune of leverage ratios.
Finally, using JordA (2005) projection technique to confirm financial accelerator
in aggregate data. However while applying the same technique to the panel data
for US banking system. I find that for the largest decile of banks, loan volume do
not go above equilibrium level on positive TFP shock. However all other banks see
loan growth when productivity improves in the economy. Finally, the NIM for large
banks never see any positive jump on productivity rise and stays below normal.
Global financial crisis of 2007-2012 prompted regulatory bodies to protect the
global financial system through preventing the failure of systemically important fi-
nancial institutions (SIFI), or, if one does fail, limiting the adverse effects of its
failure. Basel III regulations specifically target the large banks and place a higher
bank capital requirements and capital surcharges for systemically important banks.
While the large banks are constrained by more stringent regulatory requirements,
they have historically received support from the government to avoid failure. Thus
depositors perceive these banks as safe and the SIFIs can issue deposits at a lower
cost unlike banks in other size deciles (Acharya and Mora (2012), Acharya et al.
(2010)).
The seminal papers by Bernanke et al. (1994), Bernanke et al. (1999) intro-
duced financial frictions in credit markets to the macroeconomic modeling. In their
"financial accelerator" model, the credit market frictions amplifies and propagates
aggregate productivity shock to the economy. In their model and the subsequent
literature in this area exploits various financial friction on the credit side to show
how adverse shocks to the economy may be amplified by worsening financial mar-
ket conditions. The line of inquiry pursued by several recent papers by Gertler
and Kiyotaki (2010), Gertler and Karadi (2011), Gertler et al. (2013),Gertler and
13
Kiyotaki (2015),Gertler et al. (2016) introduced realistic banks and internal moral
hazard problem to examine it's role in several aspects of financial crisis. However,
the common assumption in these papers are bank's are constrained tightly by capi-
tal requirements. Since financial intermediaries use a high leverage ratio, sometimes
in excess of 10, a tightly constrained bank would amplify any shock, both positive
and negative, to its net work by the same factor to its loan origination. This core
mechanism is the cornerstone of the entire financial friction literature.
The use of high leverage ratio in banking sector is an empirical observation and
it's rationale has received a lot of academic interest. The argument in favor of high
leverage posits that a profit maximizing bank can diversify their lending portfolio
to reduce risk and thus can produce safe deposit at a lower cost that is used by
households that do not participate any other financial market and uses bank deposits
to transfer their savings inter temporally. Thus DeAngelo and Stulz (2015), Hanson
et al. (2015), Hart and Zingales (2015) and a set of influential literature argues that
high leverage is optimal for banks. On the other hand, Admati and Hellwig (2014),
Admati et al. (2018) calls attention to the financial distress cost of high leverage and
argue that a safer banking system needs much higher capital ratio. Also, empirically
in a cross-country study Gambacorta and Shin (2018) shows that 1 percentage point
increase in the bank equity-to-total assets ratio is associated with a four basis point
reduction in the cost of debt financing and with a 0.6 percentage point increase in
annual loan growth. The common assumption in this literature is that banks lever up
as much as possible within regulatory prescription because doing so is economically
profitable.
However, the data contradicts the above assumptions. Recent empirical evidence
by Gropp and Heider (2010) shows banks maintain a high level of discretionary
capital over and above the minimum capital requirement. The discretionary capital
14
is higher than the buffers that banks hold to insure against falling below the minimum
capital requirement. They show that the standard cross-sectional determinants of
non-financial firm's capital structures can explain capital structure of large, publicly
traded banks in the US and Europe, except for banks close to the minimum capital
requirement. SIFIs are the ones who hold capital close to the minimum required
level.
The paper rest of the paper is organized as follows: the second part is a selective
literature review of the most relevant papers on financial friction and bank sizes, the
third section points to our data sources, the fourth section shows how banks differ
in cross section, the fifth section examines if banks are truly constrained by capital
requirements, the sixth section examines if the propagation of productivity shock is
different across different bank sizes and the final part concludes the paper.
2 Literature Review
The first few papers that introduced financial friction in the spirit of costly state
verification model of Townsend (1979) into the standard infinite horizon framework
was the seminal papers by Carlstrom and Fuerst (1997), and Bernanke et al. (1999).
These models focus on credit market friction, where firms have to pay external finance
premium for borrowing from credit market. The premium increases as firms use
more leverage and the credit market frictions amplifies and propagates aggregate
productivity shock to the economy.
Collateral constraint is alternative approach to the external finance premium
driven financial accelerator mechanism based on the original idea of Hart and Moore
(1994). The seminal paper by Kiyotaki and Moore (1997) posits that the creditwor-
thiness of the borrower and availability of collateral affects amount of loans and thus
15
propagate and amplify productivity shocks.
These models of financial frictions didn't have an explicit role for banks. The
friction was used on the demand side of credit and financial contracts are arranged
directly in the financial market under the known form of a contract for the acquisition
of external funding.
The DSGE literature introducing a banking sector explicitly taking into account
information asymmetry inside a bank and several features of its balance sheet hap-
pened only after the financial crisis of 2007-2012. The current research on macroeco-
nomic models with a banking sector examining the role of banks and its behavior are
a series of seminal contribution by Goodfriend and McCallum (2007), Gertler and
Karadi (2011), Gertler and Kiyotaki (2010), Gertler et al. (2013), Gertler and Karadi
(2015), Gertler and Kiyotaki (2015), Gertler et al. (2016). These papers add various
financial frictions to account for differences between the interbank interest rate and
other short rates, role of monetary policy actions, monetary policy transmissions and
Fed's unconventional policy instruments during the financial crisis.
The common assumption in all of these papers is the fact that banks are con-
strained by capital requirements arising from either internal moral hazard problems
or external regulatory requirement. The role of bank size in determining bank capi-
tal structure, the empirical evidence of discretionary capital in excess of regulatory
requirement or differential regulatory treatment towards bank of different sizes has
not received any attention in the current literature.
The first theoretical paper to do so is by Davila and Walther (2017). The authors
examine how large and small banks make funding decisions when the government
provides system-wide bailouts to the financial sector. They show that bank size,
purely on strategic grounds, is a key determinant of banks' leverage choices, even
when bailout policies treat large and small banks symmetrically. In their article, large
16
banks always take on more leverage than small banks because they internalize that
their decisions directly affect the government's optimal bailout policy. In equilibrium,
small banks also choose strictly higher borrowing when large banks are present, since
banks' leverage choices are strategic complements. Overall, the presence of large
banks increases aggregate leverage and the magnitude of bailouts. They show that
the optimal ex-ante regulation features size-dependent policies that disproportionally
restrict the leverage choices of large banks as increase in the share of assets held by
the largest bank is associated with a higher aggregate debt-to-asset ratios.
Optimal bank capital structure received a lot of theoretical attention especially
in the spirit of bank run model by Diamond and Dybvig (1983). However, there is
scant empirical evidence for the effect of size on bank capital structure decisions.
The only empirical paper which touches upon this question is by Gropp and Heider
(2010). The authors show, contrary to current theoretical understanding, mispriced
deposit insurance and capital regulation are of second-order importance in determin-
ing bank capital structure of large U.S. and European banks during 1991 to 2004.
They show that standard cross-sectional determinants of non-financial firms' leverage
carry over to banks, except for banks whose capital ratio is close to the regulatory
minimum. Consistent with a reduced role of deposit insurance. The paper docu-
ments a shift in banks' liability structure away from deposits towards non-deposit
liabilities. Also the authors find that unobserved time-invariant bank fixed-effects
are ultimately the most important determinant of bank capital structures and that
bank leverage converges to bank specific, time-invariant targets. This shows that un-
derlying bank-specific characteristics affect the capital structure decisions which may
not be explained by the current modeling techniques and calls for a re-evaluation of
what I know about optimal level of bank capital.
17
3 Data source
I have used various publicly available data sources as described below.
Bank data. The bank data is from U.S. Call Reports provided by the Federal
Reserve Bank of Chicago from WRDS Bank Regulatory database. I make several
changes to make the definition of data consistent across time as suggested by
Drechsler et al. (2017). The data is from January 1994 to December 2013. It
contains quarterly data on the income statements and balance sheets of all U.S.
commercial banks.
Deposit holdings. The data on deposit quantities is from the Federal Deposit
Insurance Corporation (FDIC) available using WRDS Bank Regulatory database.
The data covers the universe of U.S. bank branches at an annual frequency from
June 1994 to June 2014.
Interest Rates. The interest rate data is from WRDS RATES database is based
upon the Federal Reserve Board's H.15 release that contains selected interest rates
for U.S. Treasuries and private money market and capital market instruments. All
rates are reported in annual terms. Daily figures are for Business days and Monthly
figures are averages of Business days unless otherwise noted. The data is from
January 1994 to December 2013.
Macroeconomic Data. The macroeconomic data is retrieved from Valerie
Ramey's website.The author has made available the data and codes used in the
paper Ramey (2016). The data contains quarterly macroeconomic data between
1947Q1 and 2015Q4. It contains nominal and real data on GDP, nominal
consumption, total investment, nonresidential investment, hours worked, output
and output per hour in private business, relative price of investment goods, total
population, interest rates on BAA bond, 10 year Treasury bond, 3-month Treasury
18
bill and stock prices, from Shiller and Fama-French among several other variables.
TFP Data. Total factor productivity data is created by John Fernald and
obtained from Federal Reserve Bank of San Francisco. Fernald (2014) shows the
methodology for creating a real-time, quarterly series on total factor productivity
(TFP) for the U.S. business sector, adjusted for variations in factor utilization -
labor effort and capital's workweek. I use quarterly data between 1947Q1 and
2015Q4 for estimating economy wide impulse response and data between 1994Q1
and 2013Q4 to merge with bank data.
4 Bank Characteristics and Summary Statistics
Prudent management of banks and other financial institutions is important as depos-
itors are generally small unsophisticated investors each typically maintaining a single
banking relationship. On failure of a scheduled commercial bank, the government
(through FDIC) compensates the depositors and is liable for the cost. Thus it is
important for regulators these institutions maintain enough capital to avoid moral
hazard problems and ensure continuation even under adverse economic condition.
Internationally the bank capital requirements is set by the Basel Accords. The
Basel framework mandates the banks to maintain a minimum capital adequacy ratio.
The capital adequacy ratio is the minimum required percentage of a bank's capital
to its risk-weighted assets. In the United States, a bank holding company must have
a Tier 1 capital ratio of at least 4%, a combined Tier 1 and Tier 2 capital ratio of
at least 8%, and a leverage ratio of at least 4%. These requirements ensures that
the banks cannot take excess leverage and become insolvent. In addition to capital
requirements which govern the liability side of the balance sheet, there are reserve
requirements, which govern the assets side of a bank's balance sheet. The reserve
19
requirement is minimum required percentage of cash or highly-liquid assets to its
total assets.
In addition to Basel accords, all US banking institutions above the size of 10
billion USD are subjected to stress test since the financial crises in 2007-2009 as part
of the Comprehensive Capital Analysis and Review (CCAR) and the Dodd-Frank
Wall Street Reform and Consumer Protection Act ("Dodd-Frank Act"). A bank stress
test is a simulation of the balance sheet of that institution under baseline, adverse
and severely adverse macroeconomic scenarios. Each scenario includes economic
variables, including macroeconomic activity, unemployment, exchange rates, prices,
incomes and interest rates. These ensure that financial institutions have well-defined
and forward-looking capital planning processes that account for their unique risks
and sufficient capital to continue operations through times of economic and financial
stress. Also using CCAR Federal Reserve restricts bank's capital distributions, such
as dividend payments or stock repurchases.
I use bank data from 1983Q1 to 2013Q4 and calculate several capital structure
ratios for these banks. I create 10 quantiles of banks according to their asset size
in each quarter. Next I use the time-average of these ratios in each decile to find
the decile average that is plotted. Given 1.6 million observations across bank and
quarters, it is expected that simple averaging will show no pattern across deciles.
However, the following summary statistics and graphs are really revealing.
The figure 1 graphs average equity to asset by size decile show a steep downward
pattern across deciles. The banks in smallest size deciles hold 13% equity against
the total asset. The largest banks maintain an average equity to asset ratio of 0.08.
8% is also the minimum capital requirement according to BASEL III. The decile
averages are perfectly monotonic and downward sloping between the lowest and the
highest deciles. Hence it is evident that banks in smallest size deciles hold equity in
20
Average Equity to Asset by Size Decile
't
0
0e
0
Cr
0U0
00
0 2 4 6 8 10
10 quantiles of assets
Time Period: 1983Q1 to 2013Q4
Figure 1: Average Equity to Asset Ratio by Size Decile
excess of minimum capital requirement. This makes sense as the smallest banks are
also the riskiest and the optimal capital structure would require them to trade off
between costs of bank distress and the liquidity creation. This explanation follows
from the seminal paper by Diamond and Rajan (2000).
The figure 2 graphs average book leverage ratio by size decile, average interest
income and interest expense and average net interest margin. The relationship for
average book leverage ratio is monotonically increasing in size. This is expected
given its negatively related to equity to asset ratio. However the other relationships
are not monotonic and shows interesting pattern. The average interest income varies
between 6.8% and 7.5%. The interest income is calculated as total interest income
over total assets. The income increases up to decile 3 and then steadily decile until it
suddenly jumps for the largest bank decile. The difference is statistically significant.
21
Average Book Leverage Ratio by Size Decile Average Interest Income by Size Decile
0
04*
S
0
0
S
S
.2
S 0
1 -C, 04
S
00 0
0
0
C9
0 2 4 8 10 ,0 2 6 8 10
10 quantiles of assets 10 quantiles of assets
Time Period: 198301 to 201304 Time Period: 198301 to 201304
Average Interest Expense by Size Decile Average Net Interest Margin by Size Decile
Coj
0
0 l
CO
Coi
0
001 0 0 0
.0.
0
N
0
S
0
*6*
0~
0 4 6 S 10 0 2 4 6 8 10
10 quantiles of assets 10 quantiles of assets
Time Period: 1983Q1 to 201304 Time Period: 198301 to 2013Q4
Figure 2: Average Leverage, Interest Income, Interest Expense and NIM by Size
Decile
22
The average interest expense is calculated as the total interest expense over total
liabilities. The average interest expense increases from decile-1 to decile-6 and then
falls. The range of interest expense is between 2.4% and 2.8%. Due to outlier nature
of the interest income, the net interest margin (NIM) is the highest for the largest
banks. The relationship between net interest margin and size is negative and NIM
falls steadily as size increases for all the other deciles.
The figure 3 graphs several other capital structure ratios by size decile. Average
loan to asset ratio is monotonically increasing in size and varies between 0.5 and 0.6.
Average deposit to asset ratio doesn't show a clear pattern and varies between 0.84
and 0.88. Average non-deposit liabilities is monotonically increasing in size. These
observation is in line with Gropp and Heider (2010) which shows that the largest
banks use higher non-deposit liabilities compared to smaller banks. The average
non-deposit liabilities for banks in decile-10 is 0.1 whereas banks in lower deciles
raise non-deposit liabilities an average of 0.02 times their total assets.
The Tier 1 capital ratio is the ratio of a bank's core equity capital to its total
risk-weighted assets (RWA). Risk-weighted assets are the total of all assets held
by the bank weighted by credit risk, components for market risk and operational
risk. Tier-1 capital to risk-weighted asset is monotonically decreasing in size. This
is an important observation. The largest banks maintain the lowest Tier-1 capital
to risk-weighted asset even though the BASEL-III mandates minimum total loss-
absorbing capacity (TLAC) requirement to be the highest for these banks. The
lowest size banks maintain a Tier-1 capital ratio close to 25% where as the largest
banks maintain a ratio close to 15%. However, even then the ratio is higher than the
TLAC requirements for the largest banks.
Table 1 provides summary statistics for all banks in the entire panel data (1983Q1
to 2013Q4). We find that the mean of book leverage is 0.9 with a standard deviation
23
Average Loan to Asset by Size Decile Tier-i Capital to Risk Weighted Asset by Size Decile
0
S
e S
Co
CO -
0 2 4 6 10 0 2 4 8 10
10 quantiles of assets 10 quantiles of assets
Time Period: 198301 to 201304 Time Period: 199601 to 2013Q4
Average Deposit to Asset by Size Decile Average Non-deposit Liabilities to Asset by Size Decile
0
. S g
S
0
co
0*
crj
(D-
0e
2.-
S
0-
8 10 0 2 6 8 1o
10 quantiles of assets 10 quantiles of assets
Time Period: 198301 to 201304 Time Period: 198301 to 2013Q4
Figure 3: Average Loan to Asset, Tier-1 Capital to Risk Weighted Asset, Average
Deposit to Asset and Average Non-Deposit Liabilities to A byt Size Decile
sse
24
of 0.039. There is a variability in book leverage across banks. Interestingly, Tier-1
capital to RWA ratio has a mean of 0.169 and a SD of 0.109. The variance in Tier-1
capital ratio is especially large which shows banks are likely not constrained by the
regulatory requirements on this ratio.
Table 2 provides summary statistics for the subset of banks in the highest decile
and the highest percentile (1983Q1 to 2013Q4). I find that the mean of book leverage
is 0.9 with a standard deviation of 0.039. There is a variability in book leverage across
banks. Interestingly, Tier-1 capital to RWA ratio has a mean of 0.169 and a SD of
0.109. The variance in Tier-1 capital ratio is especially large which shows banks are
likely not constrained by the regulatory requirements on this ratio. Looking at the
highest deciles of banks it is clear interest earned on loans. book leverage, loan to
asset ratio, NIM is higher than average, whereas equity to asset ratio and Tier-1
capital to risk weighted capital is lower.
5 Investment Cash Flow Sensitivity of Banks
I show in the previous section that, on average, large banks maintain lower equity
to asset ratio and Tier 1 capital to risk weighted asset ratio. As these ratios are
indeed close to the minimum capital requirements, equity remaining unchanged large
banks cannot increase their deposits and loans without falling below the capital
requirements. Thus only when bank capital increases, banks can issue new deposit
and make new loans which were otherwise not possible. Therefore when capital
requirements constraint binds, the deposits and loans should be more sensitive to
internal cash flow generated by the banks. As internal cash flows increase bank
capital, banks would be able borrow more by issuing more deposits. They would
also deploy these two sources of funds, net profit and additional borrowing, to lend
25
Obs Mean SD Mean Max
Ratios:
Interest Earned on Loans 1,605,645 0.070 0.046 0.000 0.145
Interest Expense on Deposits 1,609,490 0.027 0.023 0.000 0.078
Book Leverage 1,609,476 0.903 0.039 0.693 0.960
Loan to Asset Ratio 1,609,474 0.567 0.150 0.143 0.875
Equity to Asset Ratio 1,609,474 0.088 0.045 0.071 0.254
Deposit to Asset Ratio 1,609,476 0.862 0.075 0.448 0.944
Non-deposit to Asset Ratio 1,609,476 0.039 0.056 0.000 0.333
Tieri Capital to Risk Weighted Capital 548,318 0.169 0.109 0.073 0.849
Net Interest Margin to Asset Ratio 1,605,640 0.043 0.029 0.000 0.126
Balanace Sheet:
Assets (USD Million) 1,609,479 186.1 569.3 3.5 4634.0
Net Interest Margin (USD Million) 1,148,463 2.3 7.1 0.0 58.1
Deposits (USD Million) 1,609,478 142.6 392.7 2.6 3123.0
Loans (USD Million) 1,609,476 114.2 355.3 1.1 2878.0
Equity (USD Million) 1,609,478 16.3 47.7 0.3 386.6
Table 1: Summary Statistics for All Banks in the entire Panel Data (1983Q1 to
2013Q4)
26
Highest Decile Highest PercentileHighest Decile Highest Percentile
Obs Mean SD Obs Mean SD
Ratios:
Interest Earned on Loans 160,708 0.075 0.042 16014 0.092 0.030
Interest Expense on Deposits 160,871 0.027 0.023 16014 0.027 0.024
Book Leverage 160,871 0.919 0.030 16014 0.926 0.032
Loan to Asset Ratio 160,871 0.607 0.140 16014 0.602 0.142
Equity to Asset Ratio 160,871 0.082 0.030 16014 0.077 0.024
Deposit to Asset Ratio 160,871 0.804 0.116 16014 0.655 0.137
Non-deposit to Asset Ratio 160,871 0.109 0.095 16014 0.237 0.096
Tieri Capital to Risk Weighted Capital 54,800 0.127 0.069 5449 0.108 0.052
Net Interest Margin to Asset Ratio 160,708 0.048 0.031 16014 0.064 0.034
Balanace Sheet:
Assets (USD Million) 160,871 1,212.0 1,413.8 16014 4,017.9 1,112.6
Net Interest Margin (USD Million) 114,791 15.4 17.6 11422 53.1 11.3
Deposits (USD Million) 160,871 872.3 948.5 16014 2,655.7 789.4
Loans (USD Million) 160,871 747.3 884.4 16014 2,432.6 763.8
Equity (USD Million) 160,871 99.4 119.0 16014 309.7 112.9
Table 2: Summary Statistics for Banks in Highest Decile and Highest Percentile in
the entire Panel Data (1983Q1 to 2013Q4)
27
more.
In this section, I establish the effect of regulatory constraint on bank lending and
deposit taking behavior controlling for lending opportunities. Our goal is to show
that the large banks are more constrained on their lending and deposit taking than
smaller banks due to the binding capital ratio regulations. The empirical strategy is
to show that both deposit cash flow sensitivity and lending cash flow sensitivity are
higher for larger banks.
This strategy is similar to Fazzari et al. (1988) investment cash flow sensitivity
analysis for manufacturing firms. Due to asymmetric information in credit markets,
more constrained firms use internal capital as the source of funds for investments
and has a higher investment cash flow sensitivity controlling for lending opportuni-
ties. They use three different proxies for lending opportunities in different empirical
specifications - Tobin's Q, the market value of the assets to the book value; sales ac-
celerator, the fluctuations in sales relative to firm's equity; and a combined measure
of output and the cost of capital.
In our case, I use two different proxies for bank lending opportunities in different
empirical specifications. The first proxy is the spread between 10 year constant
maturity government bond rate and 3 months treasury bill rate. As banks issue
shorter maturity liabilities to fund longer maturity loans, the spread is reflective
of lending opportunities available to all the banks but it does not reveal the cross
sectional heterogeneity across banks. The second proxy is the net interest margin
(NIM), defined as the ratio of net interest income to total assets less net interest
expenses on deposits. NIM determines the profitability of bank lending and reflects
the cross sectional heterogeneity across banks.
28
I estimate the following models in the panel data:
10
(1)
it ai + t + E LjSize-Decilej x +- K yXit + 
it 
j= )it
where it represents loans for bank i during period t in the first specification,
deposits in the second specification. CFjt represents the net profit earned by bank i
during period t. Both right and left hand side variables are scaled by bank capital
Kit. I interact this variable with 10 size deciles for each bank. Size decile indicates
the decile bank i belongs to according to bank asset size at time t. X represents the
control for lending opportunity. The first set of regressions use bond spread as the
proxy which has no cross sectional heterogeneity at each time period t. aj are bank
fixed effects and 6t are time fixed effects. I cluster standard errors at the bank level.
To ensure that the regression results are not driven by the size of the banks, I
scale both the dependent variable using bank capital. The same can be done using
bank assets as scaling factor. But I stay as close to Fazzari et al. (1988) specification
as possible. To facilitate interpretation, I scale the dependent variables by 100.
Hypotheses. I expect to see similar coefficients on loan and deposit growth in
response to increasing net profit across different size deciles if all banks are equally
capital constraint. I do not expect to see any discernible pattern on the coefficients
when sorted on the basis of size.
Results. The results are presented in the following tables.
In table 3 I report the regression results for quarterly loans and deposits on
net profit controlling for bond spread as investment opportunity. The first set of
regressions are contemporaneous. In the next set of specifications I use next quarter
lending and borrowing as the dependent variable.The stars on the coefficients show
levels of significance: for p-value is less than 0.1, it is flagged with one star (*), for
29
p-value is less than 0.05, it is flagged with two stars (**) and finally a p-value is less
than 0.01, it is flagged with three stars (***). Standard errors of the coefficients are
shown in the parentheses.
In table 4 I report the regression results for quarterly loans and deposits on net
profit controlling for NIM as investment opportunity. Like previous table results for
contemporaneous and next quarter lending and borrowing is reported. I follow the
same convention for statistical significance.
According to table 3, for the smallest banks in decile-1, every dollar increase
in net profit increases the loan issuance by 38 cents contemporaneously and issue
51 cents more deposit. A dollar increase in net profit in the previous period, also
contributes to 32 cents increase in loan and 43 cents increase in deposits in the
next period. All the coefficients are statistically significant at the level of 1%. This
implies the banks in this decile are constrained by the capital requirements. As the
net profit increases and the bank capital is improved after paying dividends, they can
issue more deposit and thus lend more. However, for every dollar increase in capital,
banks can increase the deposits by close to 10 times. Here I see that the increase
in deposits is less than a dollar on the net profit. The same numbers increase to 51
cents per dollar in loan and 56 cents per dollar in deposits for banks in 10th decile.
If I look at the coefficients for contemporaneous loan on size in column (1) of
table 3 are monotonically increasing. This means larger the bank issue more loan
contemporaneously as their net profit number increases. The difference between
10th decile and 1st decile is statistically significant in t-test. The result is exactly
similar in column (3) for next quarter loan issuance. This is an indirect evidence that
larger banks are more capital constrained. As the bank capitals increase larger banks
increase their loan volume more than smaller banks both contemporaneously and in
the next period. These results cannot be explained by investment opportunities as I
30
control for investment opportunities using 10-year bond spread.
In the column (2), the coefficients on contemporaneous deposits increase with
size up to 8th decile and then decreases in decile 9 and 10. This shows as profits
improve the bank capital, larger banks choose to issue more deposits, however banks
in largest deciles can also issue more non-deposit liabilities. This similar results are
obtained for next quarter deposits as well in column (4). All the coefficients are
statistically significant at 1% level of significance.
In the previous table, investment opportunity is controlled by 10-year bond spread
which is same in the cross-section. To distinguish between differential investment
opportunities available to different banks, I construct NIM variable for each bank.
In table 4, I control the investment opportunity using NIM-variable. In column
(1) and column (3), the results are similar to the results obstained in 3. However,
the coeffficents on loan growth across size in both contemporaneous and one-period
in the future are not monotonically increasing anymore. Possibly because NIM is
so high in the last two deciles, its coefficient captures some of the increase in loan
growth due to increase in bank capital. All the coefficients are statistically significant
at 1% level of significance.
In column (2) and column (4), the coefficients on deposits show similar pattern
that of column (1) and (3). As size increases, the deposit growth increases with size
from 56 cents per dollar in decile-1 to 65 cents per dollar in decile-8. However, in
the largest two deciles, banks increasingly rely on non-deposit liabilities, hence the
coefficients dip. However all the coefficients are statistically significant at 1% level
of significance.
There are several ways, this estimates can be improved. Using post-tax and
post-dividend net profit as independent variable would directly make the coefficients
indicative of tightness of capital constraint. The coefficients will likely be larger than
31
reported. However, bank fixed effects should be able to absorb some of the differences
if I assume the banks keep the dividends almost constant.
This results are strongly suggestive of the fact that smaller banks are far less
constrained than the larger banks. None of the banks increase the loan dollar for
dollar contemporaneously or issue deposit in the multiples of leverage ratio as the
capital increases. This is in stark contrast with macroeconomic models which use a
mechanism that amplifies shocks to net worth in the multiples of leverage ratio and
propagate that shock to bank's loan portfolio.
6 Estimating Impulse Response in the Data
Financial accelerator model shows a mechanism by which adverse productivity shocks
to the economy gets amplified by the financial markets. Seminal papers by Bernanke
et al. (1999), Carlstrom and Fuerst (1997) posits that the external finance premium in
the credit market, defined by the difference between the cost of external borrowing
and the opportunity cost of internal funds, has a negative relationship with the
productivity shock. The external finance premium in the economy is difficult to
quantify. The usual proxy is the difference between the interest rate on loans and the
marginal cost of funds for banks, i.e., bank's price-cost margin. The usual assumption
is that bank's cost reflects the opportunity cost of firms internal capital and bank's
price is the interest rate on loans made to the firms.
There is some empirical evidence of countercyclical price-cost margins in the
US and European banking. Aliaga-Diaz and Olivero (2010) uses VAR analysis to
measure the comovement between bank margins and output in a panel data for the
US banks and show margins are countercyclical with respect to GDP and bank loans
after controlling for monetary policy and credit risk. Altunba et al. (2016) reports
32
countercyclicality of banks' price - cost margins in a panel dataset of European banks.
However, given our previous exposition on how bank size has an effect on its
capital structure decisions, it's funding cost and return on lending opportunities, I
need to re-examine this counter-cyclical price-cost margin across different bank sizes.
This is particularly relevant for policy implications that talk about even stricter
capital requirements as policy tool. If all banks already maintain capital in excess of
the regulatory minimum and do not adjust the shock to its net worth immediately
by passing it to the loan portfolio, the policy tool may be ineffective or in some cases,
counter productive.
I do two experiments in the following sections, first we look at the aggregate
economy and examine banks role in financial acceleration. In the next set of tests
I look at impulse response of several price-cost and loan portfolio variables to the
capitalization-adjusted technology factor productivity shock (TFP) across different
bank sizes.
6.1 IRF in Aggregate Time Series Data
In this section I estimate the impulse response to a TFP shock in aggregate time
series data on a number of endogenous variables of interest. The real variables
include output (GDP), consumption and total credit to nonfinancial sectors. The
price variables include short term interest rate (the 3 month TBill rate), FED funds
rate, and three measures of price-cost margin - net interest margin, the 10-year
constant maturity bond spread over the 3 months TBill rate and the BAA corporate
bond spread. Although our interest lies in estimating impulse response of price-cost
margin, I show the results in conjuction with other output variables to point out
cyclicality of the response.
(1) (2) (3) (4)
Cash Flow Loant Depositst Loant+1 Depositst+1
Size Decile 1 38.8134*** 51.4505*** 32.2093*** 43.3450***
[0.378] [0.492] [0.402] [0.506]
Size Decile 2 42.7859*** 55.6187*** 36.3619*** 47.4906***
[0.293] [0.370] [0.313] [0.381]
Size Decile 3 45.0730*** 57.3380*** 38.7090*** 49.1782***
[0.260] [0.334] [0.276] [0.340]
Size Decile 4 46.4626*** 58.5742*** 40.1408*** 50.2897***
[0.2431 [0.319] [0.259] [0.325]
Size Decile 5 47.7250*** 59.7490*** 41.5102*** 51.5097***
[0.239] [0.3171 [0.251] [0.321]
Size Decile 6 48.6486*** 60.4535*** 42.3855*** 52.0600***
[0.247] [0.328] [0.258] [0.330]
Size Decile 7 49.5104*** 61.0550*** 43.2405*** 52.6275***
[0.255] [0.348] [0.266] [0.346]
Size Decile 8 50.1067*** 60.9780*** 43.8443*** 52.4072***
[0.286] [0.404] [0.294] [0.396]
Size Decile 9 50.6489*** 59.9968*** 44.4026*** 51.4170***
[0.316] [0.466] [0.327] [0.451]
Size Decile 10 50.6553*** 56.1440*** 44.4026*** 47.6681***
[0.414] [0.724] [0.427] [0.668]
Observations 1148277 1148280 1129387 1129389
Bank FE Yes Yes Yes Yes
Quarter FE Yes Yes Yes Yes
R-squared 0.821 0.802 0.773 0.747
Adjusted R-Sq 0.818 0.799 0.77 0.743
Table 3: Cash Flow Sensitivity of Bank Loans and Deposits Controlling for 10-Year
Bond Spread
34
(1) (2) (3) (4)
Cash Flow Loant Depositst Loant+i Depositst+1
Size Decile 1 44.5197*** 56.7030*** 37.6087*** 48.2098***
[0.339] [0.458] [0.376] [0.485]
Size Decile 2 48.2264*** 60.6140*** 41.5041*** 52.1135***
[0.268] [0.351] [0.301] [0.376]
Size Decile 3 50.5478*** 62.3630*** 43.8719*** 53.8164***
[0.241] [0.320] [0.271] [0.341]
Size Decile 4 51.9104*** 63.5829*** 45.2744*** 54.9129***
[0.227] [0.3081 [0.255] [0.328]
Size Decile 5 53.0896*** 64.6855*** 46.5636*** 56.0640***
[0.221] [0.304] [0.245] [0.320]
Size Decile 6 53.9038*** 65.2851*** 47.3357*** 56.5228***
[0.227] [0.312] [0.250] [0.326]
Size Decile 7 54.5855*** 65.7353*** 48.0178*** 56.9440***
[0.232] [0.328] [0.253] [0.338]
Size Decile 8 55.0314*** 65.5429*** 48.4827*** 56.6197***
[0.254] [0.375] [0.273] [0.378]
Size Decile 9 55.3233*** 64.3614*** 48.8120*** 55.4484***
[0.282] [0.439] [0.303] [0.434]
Size Decile 10 55.1602*** 60.4152*** 48.6583*** 51.6201***
[0.371] [0.709] [0.393] [0.658]
Observations 1,144,662 1,144,663 1,125,898 1,125,899
Bank FE Yes Yes Yes Yes
Quarter FE Yes Yes Yes Yes
R-squared 0.838 0.811 0.787 0.753
Adjusted R-Sq 0.835 0.808 0.784 0.749
Table 4: Cash Flow Sensitivity of Bank Loans and Deposits Controlling for NIM
35
Several data measures of TFP shocks are constructed in the literature - Fran-
cis et al. (2014); Justiniano et al. (2011); and Fernald (2014) are the three com-
mon neutral technology shocks used in Ramey (2016). I consistently use Fernald
utilization-adjusted TFP shocks for our analysis. All three measures are highly cor-
related, but only Fernald shocks are publicly available, real-time updated, quarterly
time-series data that adjusts for variations in factor utilization - labor effort and
capital's workweek.
I use Jordi (2005) local projection method for estimating the impulse response.
This lets us compute impulse responses without specification and estimation of the
underlying multivariate dynamic system. I estimate local projections at each period
of interest as compared to extrapolating VAR estimates into increasingly distant
horizons. The impulse response of Y at horizon t+ h, is estimated using the following
single regression.
Yt+h /h - 01hEat + 1hXt_1 + Quadratic trend + Ut+h (2)
where Eat is data measure of productivity shock, Fernald-utilization adjusted shocks
in our case. Controls Xt_ 1 include two lags each of the shock to absorb any serial auto
correlation, log real GDP per capital, log labor productivity and other dependent
variables chosen for each Y variable as likely forecasters of Yt+h given time t - 1
information. Quadratic trend removes both linear and quadratic trend present in
the response variable. The coefficient #1h estimates the response of the Y variable
to a 1% productivity shock h quarters after impact, i.e., /1h -- y+Yh
a-Eat
Results. In our estimation using this data and Jordh (2005) estimation tech-
nique, impulse response to the Fernald's utilization-adjusted TFP shock results in
the usual positive jump as expected for three real variables. GDP, consumption
36
CO,
0-
0
cm,
0-
0
0-
0N
0-
0- 0-0
0- 0
0-
110 15 20 0 5 1,0 15 20
Quarters Quarters
Log of Real GOP Log of Real Consumption
0. N.
C,l 
0-
0D
0-
Log 1of 1 2' 6 5 10 1s 20
Quarters Quarters
Log of Real Credit to Nonfinancial Sectoi Fed Funds Rate
Figure 4: Effects of TFP shock, Jorda local projection, on real variables
The figure shows responses of real GDP, real consumption, real credit to nonfinancial sector
and the FED Funds rate on Fernald's utilization adjusted TFP growth. Light gray bands
are 90% confidence bands.
37
0D
LO
0-
0-
2O
CO)
0 5 10 15 20 5 10 15 20
Quarters Quarters
Net Interest Margin 3-Month Treasury Bill
LO U)
0-
0O 0
U)
0 5 10 15 20 0 5 10 15 20
Quarters Quarters
10-Year Treasury Constant Maturity Spr ad BAA Corporate bond Spread
Figure 5: Effects of TFP shock, Jorda local- projection, on price cost margin
The figure shows responses of short term interest rate (the 3 month TBill rate), net interest
margin, the 10-year constant maturity bond spread over the 3 months TBill rate and the
BAA corporate bond spread on Fernald's utilization adjusted TFP growth. Light gray
bands are 90% confidence bands.
38
shows a delayed jump and a hump-shaped response. Similarly, the credit to non-
financial sector does not jump immediately and keep on rising till 10-quarters and
then returns to normal around 16-20 quarters.
I now look at price-cost margin data using this estimation technique. There is
no empirical literature that uses this technique on aggregate data. Aliaga-Diaz and
Olivero (2010) uses VAR analysis to measure the comovement between bank margins
and output in a panel data for the US banks and show margins are countercyclical
with respect to GDP. VAR estimates extrapolate coefficients into increasingly distant
horizons.
I use three proxies for price-cost are NIM average of all banks, 10-year treasury
constant maturity spread with 3-month treasury bill, BAA bond spread against 3-
month treasury bill. I find that consistent with the empirical evidences, net interest
margin falls immediately and is statistically significant. It continues below the long
run average for at least 12-15 quarters before reverting to normal around 20 quarters.
However, both the 10-year bond spread and the BAA corporate bond spread jumps
up for first 2-3 quarters and then continue falling till 10 quarters and subsequently
regress to the mean within 16 quarters.
However, the next two measures of 10-year bond spread and BAA corporate bond
spread show interesting results. These are used in the macroeconomic literature as a
measure for investment opportunity. 10-year bond spread jumps immediately when
a positive TFP shock hits at time-0 and then starts falling after 1 or 2 quarters.
Corporate bond spread is not expected from the financial accelerator model. In
response to TFP shocks, the model posits that these rates, which are representative
of the external finance premium, should fall immediately. However, in aggregate
data, this short upswing is not statistically significant. Hence, in the next section, I
re-investigate this dynamics in panel data.
39
6.2 IRF in the Panel Data for US Banking
In this section I estimate impulse response of several bank specific variables of interest
- total loans, interest earned on loans, interest paid on liabilities and average net
interest margin to Fernald's utilization adjusted TFP shock.
This estimation is done using panel data for all banks available in call reports
maintained with WRDS bank regulatory database. The data definition is made
consistent using definitions suggested by Drechsler et al. (2017). I use JordA (2005)
local projection method in panel data.
Yt+h - h + flh6at + / 2h Xt-1 - Quadratic trend + Ut+h (3)
where Eat is data measure of productivity shock, Fernald-utilization adjusted shocks
in our case. Controls Xt_ 1 include two lags each of the shock to absorb any serial
auto correlation. Quadratic trend removes both linear and quadratic trend present
in the response variable. The coefficient /1h estimates the response of the Y variable
to a 1% productivity shock h quarters after impact, i.e., =-1h2 ,+.
Ocat
To contrast between banks which are truly constrained by regulatory capital
requirements. I split the sample in two: in the first sub-sample I include all banks
except 10th size decile, the other sub-sample is the banks comprising of 10th size
decile.
Results. In our estimation using this data and JordA (2005) estimation tech-
nique, impulse response to the Fernald's utilization-adjusted TFP shock results in
the usual positive jump as expected for three real variables. GDP, consumption
shows a delayed jump and a hump-shaped response. Similarly, the credit to non-
financial sector does not jump immediately and keep on rising till 10-quarters and
then returns to normal around 16-20 quarters.
40
I now look at price-cost margin data using this estimation technique. There is
no empirical literature that uses this technique on aggregate data. Aliaga-Diaz and
Olivero (2010) uses VAR analysis to measure the comovement between bank margins
and output in a panel data for the US banks and show margins are countercyclical
with respect to GDP. VAR estimates extrapolate coefficients into increasingly distant
horizons.
Figure 6 shows the result of the impulse response in the panel data. The results
are interesting for the highest decile as it contradicts the typical assumptions in the
macroeconomic models.
For the total bank loan portfolio, in case of all banks except the highest decile,
the response is an immediate jump to 5 basis points within first 4 quarters and then
returns to normal 8-10 quarters. However, in case of the banks in highest decile, the
loan portfolio falls by 10 basis points on 1 percentage point TFP shock and for 9
quarters stay below the equilibrium level and returns to normal in quarter 10.
The results for interest earned shows an interesting pattern, for banks in first
9 deciles, the interest earned shows a initial jump of 6 basis points and falls below
normal after 4 quarters and stays below normal for at least 10 quarters. The interest
earned on loans show a lesser jump of 4 basis points and falls below normal after 4
quarters and stays below normal for at least 10 quarters.
The results for interest paid on liabilities is completely bunched together for both
sub-samples. The results show a jump of 5 basis points within first 2 quarters for a
1% TFP shock. It stays above normal for 6 quarters and falls to -5 basis points after
8 quarters.
The results for average NIM follows similar shape in both sub-samples. However,
for the first 9 deciles, average NIM jump to 2 basis points for a 1% TFP shock and for
the highest decile it falls by 2 basis points immediately. Following this both follows a
41
C1
q
014
A /
f C0
I 0M N
/
~ N'\ /\ /~J
*1~
108461  0 2 4 
Quarters Quarters
Log of total loans Interest earned on Loans
- - - (Highest Decile) - - - (Highest Decile)
NY
0
A
C4
~; \
/
/
0
0-
0
%%
6 8 10 0 2 4 8 10
Quarten Quarters
Interest paid on Liabilities Average Net Interest Margin
- - -(Highest Decile) - - - (Highest Decile)
Figure 6: Effects of TFP shock, Jorda local projection, on total loans, interest income,
interest expense and average NIM
The figure shows responses of total banks loans, interest earned on loans, interest paid on
liabilities and average net interest margin on Fernald's utilization adjusted TFP growth.
Light gray bands are 90% confidence bands.
42
downward trend for 3-4 quarters and reach -6 basis points. Then the average NIMs
reach normal in 8 quarters.
The possible explanations for these results point to the fact that banks in highest
decile are constrained by regulatory capital requirements. As a positive TFP shock
hits the economy, they cannot adjust the loan portfolio immediately. Although
interest earned jumps by 4 basis points immediately because the capital is now
more productive, the interest expense jumps by 6 basis points because investors
now demand more return for holding the capital. This results in a negative jump in
NIM. This essentially means that capital constrained banks sees a negative growth in
profits when a positive TFP shocks appear. This will result in further worsening loan
portfolio. Hence for the highest decile, positive TFP shock results in negative growth
in loan portfolio. Whereas other banks with enough slack in capital requirements
would be able to immediately capitalize on improving economy.
The interest paid on liabilities seem to be determined by movements in short
rates which bank sizes cannot influence directly. Thus the impulse responses show
no distinct pattern across sizes. However, the differential between interest earned on
loans show that smaller banks are able to issue higher return loans when the economy
improves.
These results have interesting implications for the macroeconomic prevalence for
an accelerator mechanism and policy recommendation for a higher capital require-
ments.
Firstly, the accelerator mechanism seem to work justifiably well in aggregate
data. However, it would require unconstrained banks which are able to adjust their
portfolio in response to productivity shocks. Bernanke et al. (1999) assumes banks
can arbitrarily adjust their loan portfolios and that seems to be perfect situation for
the accelerator mechanism to work.
43
The policy recommendation in the wake of financial crises of 2007-2012 was a call
for increase in minimum capital requirements. There seem to be an unexplored side
of the recommendation. As banks become more constrained, they would be slower to
adjust to the economy. The best situation seems to have a higher capital requirement
recommendations but a lower absolute bound to declare failure. There needs to be
more empirical and theoretical study to understand these implications.
7 Conclusion
This paper points out a gap in the common macroeconomic understanding that
bank's amplify shock due to balance sheet constraint put by capital regulation. Banks
are assumed to lever up to the profit-maximizing level of capital choice that trades off
risk and reward. However our empirical analysis show that for every dollar increase
in profit, banks increase their loan portfolio less than a dollar and do not immediately
adjust the deposit issuance dollar for dollar to lever up to the current leverage ratio.
The coefficients for loan growth and deposit growth are generally increasing in size
deciles. Thus largest banks are more tightly constrained and all other banks maintain
discretionary capital in excess of regulatory requirements.
I show that in cross-section across bank sizes, banks of different sizes differ sys-
temically in various capital structure parameters. The larger banks maintain signifi-
cantly lower equity, use higher leverage, higher average loan to asset and issue more
non-deposit liabilities. Whereas, the bank in largest deciles have significantly higher
average net interest income and average net interest margin.
I believe that the future macroeconomic literature needs to account for the role
of bank size in determining the role of bank capital structure in propagation and
amplification of TFP shocks. While it seems true that banks propagate and amplify
44
TFP shocks, it seems to happen differently for banks of different sizes. It is possible
to improve the infinite-horizon DSGE models by incorporating banks that maintain
discretionary capital and thus the capital constraint does not bind.
45
46
Bibliography
Viral V Acharya and Nada Mora. Are banks passive liquidity backstops? deposit
rates and flows during the 2007-2009 crisis. Technical report, National Bureau of
Economic Research, 2012.
Viral V Acharya, Joao Santos, Tanju Yorulmazer, et al. Systemic risk and deposit
insurance premiums. Citeseer, 2010.
Anat Admati and Martin Hellwig. The Bankers' New Clothes: What's Wrong with
Banking and What to Do about It-Updated Edition. Princeton University Press,
2014.
Anat R Admati, Peter M DeMarzo, Martin F Hellwig, and Paul Pfleiderer. The
leverage ratchet effect. The Journal of Finance, 73(1):145-198, 2018.
Roger Aliaga-Diaz and Maria Pfa Olivero. Is there a financial accelerator in us
banking?: Evidence from the cyclicality of banks' price-cost margins. Economics
Letters, 108(2):167-171, 2010.
Yener Altunba , Caterina Di Tommaso, and John Thornton. Is there a financial
accelerator in european banking? Finance Research Letters, 17:218-221, 2016.
Ben Bernanke, Mark Gertler, and Simon Gilchrist. The financial accelerator and the
flight to quality. Technical report, National Bureau of Economic Research, 1994.
Ben S Bernanke, Mark Gertler, and Simon Gilchrist. The financial accelerator in a
quantitative business cycle framework. Handbook of macroeconomics, 1:1341-1393,
1999.
Charles T Carlstrom and Timothy S Fuerst. Agency costs, net worth, and business
fluctuations: A computable general equilibrium analysis. The American Economic
Review, pages 893-910, 1997.
47
Eduardo Davila and Ansgar Walther. Does size matter? bailouts with large and
small banks. Technical report, National Bureau of Economic Research, 2017.
Harry DeAngelo and Ren6 M Stulz. Liquid-claim production, risk management,
and bank capital structure: Why high leverage is optimal for banks. Journal of
FinancialE conomics, 116(2):219-236, 2015.
Douglas W Diamond and Philip H Dybvig. Bank runs, deposit insurance, and liq-
uidity. Journal of political economy, 91(3):401-419, 1983.
Douglas W Diamond and Raghuram G Rajan. A theory of bank capital. The Journal
of Finance, 55(6):2431-2465, 2000.
Itamar Drechsler, Alexi Savov, and Philipp Schnabl. The deposits channel of mone-
tary policy. The Quarterly Journal of Economics, 132(4):1819-1876, 2017.
Steven Fazzari, R Glenn Hubbard, and Bruce Petersen. Investment, financing deci-
sions, and tax policy. The American Economic Review, 78(2):200-205, 1988.
John Fernald. A quarterly, utilization-adjusted series on total factor productivity.
Federal Reserve Bank of San Francisco, 2014.
Neville Francis, Michael T Owyang, Jennifer E Roush, and Riccardo DiCecio. A
flexible finite-horizon alternative to long-run restrictions with an application to
technology shocks. Review of Economics and Statistics, 96(4):638-647, 2014.
Leonardo Gambacorta and Hyun Song Shin. Why bank capital matters for monetary
policy. Journal of FinancialI ntermediation, 35:17-29, 2018.
Mark Gertler and Peter Karadi. A model of unconventional monetary policy. Journal
of monetary Economics, 58(1):17-34, 2011.
Mark Gertler and Peter Karadi. Monetary policy surprises, credit costs, and economic
activity. American Economic Journal: Macroeconomics, 7(1):44-76, 2015.
Mark Gertler and Nobuhiro Kiyotaki. Financial intermediation and credit policy
in business cycle analysis. In Handbook of monetary economics, volume 3, pages
547-599. Elsevier, 2010.
Mark Gertler and Nobuhiro Kiyotaki. Banking, liquidity, and bank runs in an infinite
horizon economy. American Economic Review, 105(7):2011-43, 2015.
48
Mark Gertler, Peter Karadi, et al. Qe 1 vs. 2 vs. 3...: A framework for analyzing
large-scale asset purchases as a monetary policy tool. international Journal of
central Banking, 9(1):5-53, 2013.
Mark Gertler, Nobuhiro Kiyotaki, and Andrea Prestipino. Wholesale banking and
bank runs in macroeconomic modeling of financial crises. In Handbook of Macroe-
conomics, volume 2, pages 1345-1425. Elsevier, 2016.
Marvin Goodfriend and Bennett T McCallum. Banking and interest rates in mone-
tary policy analysis: A quantitative exploration. Journal of Monetary Economics,
54(5):1480-1507, 2007.
Reint Gropp and Florian Heider. The determinants of bank capital structure. Review
of finance, 14(4):587-622, 2010.
Samuel G Hanson, Andrei Shleifer, Jeremy C Stein, and Robert W Vishny. Banks as
patient fixed-income investors. Journal of FinancialE conomics, 117(3):449-469,
2015.
Oliver Hart and John Moore. A theory of debt based on the inalienability of human
capital. The Quarterly Journal of Economics, 109(4):841-879, 1994.
Oliver Hart and Luigi Zingales. Liquidity and inefficient investment. Journal of the
European Economic Association, 13(5):737-769, 2015.
Oscar JordA. Estimation and inference of impulse responses by local projections.
American economic review, 95(1):161-182, 2005.
Alejandro Justiniano, Giorgio E Primiceri, and Andrea Tambalotti. Investment
shocks and the relative price of investment. Review of Economic Dynamics, 14
(1):102-121, 2011.
Nobuhiro Kiyotaki and John Moore. Credit cycles. Journal of political economy, 105
(2):211-248, 1997.
Valerie A Ramey. Macroeconomic shocks and their propagation. In Handbook of
Macroeconomics, volume 2, pages 71-162. Elsevier, 2016.
Robert M Townsend. Optimal contracts and competitive markets with costly state
verification. Journal of Economic theory, 21(2):265-293, 1979.
49