The Impact of Bank Financing on Municipalities’ Bond

Issuance and the Real Economy

Ramona Dagostino∗
University of Virginia
Darden School of Business

October 17, 2024

ABSTRACT

Do federal tax incentives for banks investing in municipal bonds support local governments
during recessions? This paper exploits a change in tax benefits for banks purchasing mu-
nicipal bonds and finds that expanding access to bank financing during recessions increases
local governments’ debt issuance and employment growth. The estimated job multiplier
is 22 jobs per million dollars of spending. There is moderate evidence of mortgage loans
being crowded out by banks’ increased holdings of municipal bonds.

JEL Classifications: H74, H72, E62

Keywords: Municipal Bonds; Tax Policy; Employment; Recession; Bank Financing;
Crowding Out; Multiplier

∗
I would like to thank Jean-Noel Barrot, Daniel Bergstresser, Murillo Campello, Joao Cocco, Igor Cunha,
Peter Feldhutter, Francisco Gomes, Daniel Green, Christopher Hennessy, Ryan Israelsen, Ivan Ivanov, Ron
Kaniel, Narayana Kocherlakota, Ralph Koijen, Stefan Lewellen, Anton Lines, Alan Moreira, Elias Papaioannou,
Anna Pavlova, Tarun Ramadorai, Helene Rey, Scott Richardson, Stephen Schaefer, Antoinette Schoar, Nathan
Seegert, Rui Silva, Vikrant Vig, Toni Whited, Emily Williams and seminar participants at UCLA Anderson,
Michigan Ross, Stanford GSB, Rochester Simon, Notre Dame Mendoza, Fed Board, Fed NY, Purdue Krannert,
Imperial, Dartmouth, HEC, UCSD, and at the SFS Cavalcade 2018, EFA 2018, WFA 2018, Gerzensee 2018, UNC
Tax Symposium 2019, and the Municipal Finance Conference for their comments on the paper. For numerous
discussions on the institutional details, I would like to thank Michael Decker, Michael Foux, William Fox, Tracy
Gordon, Michael McPherson, Shane Parker, and Missaka Warusawitharana. Finally, I thank the AQR Asset
Management Institute for generous financial support. All remaining errors are my own. Corresponding author:
Ramona Dagostino, University of Virginia, Darden School of Business, Email: dagostinor@darden.virginia.edu
 I. Introduction

During economic recessions, local governments face significant fiscal challenges due to their
dependence on volatile revenue sources combined with requirements to maintain balanced bud-
gets. A decline in revenue sources may compel local governments to make significant cuts
in spending and employment to sustain their budgets. Such cuts can adversely affect public
services, including education, public safety, and infrastructure. Moreover, reductions in local
government spending can have spillover effects on the private sector, as decreased contract work
and lower local demand further exacerbate the economic downturn. The 2007 financial crisis
and more recently the COVID-19 pandemic have demonstrated the critical need for effective
federal programs to support local governments and help mitigate the adverse effects of economic
downturns on local services, employment, and the broader economy.

In this paper, I study the effectiveness of a federal policy designed to expand local govern-
ments’ access to financing during economic downturns by increasing the tax benefits available
to banks that purchase municipal debt. As municipal bonds are typically used to finance long
term infrastructure projects, rather than to cover operating expenses, it is unclear whether this
type of federal policy is effective in providing budgetary relief to municipalities facing fiscal
challenges during a recession.1

This type of federal intervention differs from policies that involve direct grants or transfers,
as it aims to support local governments by providing a tax cut to bond investors. Differently
from grants, while banks benefit from an expanded tax incentive, the local government remains
responsible for issuing the bonds and repaying the obligations using its own resources and future
tax revenues. Similarly, this policy differs from standard loan guarantee programs, as the banks
are still exposed to the credit risk of the municipality.

Normally, banks receive federal tax advantages when holding a specific type of municipal
debt known as bank-qualified bonds. Prior to 2009, in order to receive the bank-qualified status,
municipalities could raise up to $10 million in aggregate bond issuance per year. In response
to the financial and economic crisis, the federal government temporarily raised this limit to $30
million for the years 2009 and 2010, thereby expanding the pool of municipal debt for which
banks receive tax advantages.
1
The Pew Trust reports an insightful interview on the use of debt for local governments, featuring director
and partner from Municipal Market Analytics: https://www.pewtrusts.org/en/research-and-analysis/
articles/2021/01/28/state-and-local-governments-relied-on-debt-for-budgetary-help-in-2020

1
 I start by investigating the implications of this discontinuous tax treatment for banks’ invest-
ment and local governments’ issuance decisions. I show that banks respond to this discontinuity
in tax treatment, in so far as they predominantly choose to hold bank-qualified bonds. Prior
to 2009 the bank-qualified federal tax policy effectively segmented the municipal bond market
into two groups: one where bank investors are active, consisting of issuers raising $10 million
or less in debt per year; and another where municipalities, issuing more than the $10 million
threshold, do not have access to bank financing.

Analyzing the distribution of bond issuance, I document that, prior to 2009, local govern-
ments respond to tax-induced limits on bank financing by bunching around the bank qualifica-
tion cutoff. I use a bunching methodology and estimate that issuers reduce their debt issuance
by approximately 3% on average in a year to meet the $10 million cut-off for bank investment.

To further shed light on the issuers’ response to the bank-qualification limit, I examine
the possibility that bunching is part of a dynamic borrowing strategy, whereby governments
spread their issuance across several years to remain within the $10 million limit. Analyzing
issuers’ borrowing patterns shows that bunching governments issue bonds more frequently than
non-bunching governments. However, the majority of bunching issuers do not access the bond
market in consecutive years, suggesting that dynamically adjusting debt issuances is not a
common strategy. While some local governments might strategically spread their borrowing,
the bank qualification limit still imposes significant costs for such issuers, including delayed
investment and higher issuance fees.

Did the federal policy expanding the bank qualification limit benefit local governments? To
answer this question, I focus on two set of issuers: bank-qualified issuers that bunched around
the $10 million threshold prior to 2009, and smaller bank-qualified issuers that fell below the
bunching region. The former group represents issuers for whom the limit on bank financing
was binding. The second group instead comprises municipalities whose financing needs were
significantly below the $10 million threshold and thus were not constrained by it. I aggregate
the debt issuance for these two groups at the county level and calculate the fraction of bunching
municipalities within each county. A county is defined as “treated” if it has at least one bunching
issuer. I then compare the changes in bank-qualified municipal bond issuance between treated
and control counties before and after the federal policy change.

I find that expanding federal tax cuts for banks investing in municipal bonds lead to an
economically and statistically significant increase in local government borrowing. Specifically,
I estimate an average increase in local debt issuance of around $2.7 millions for a one standard

2
deviation increase in the fraction of affected municipalities in a county.

Was the federal policy change effective in stimulating the local economy? Municipalities in
more economically depressed areas may have borrowed more than those in areas less affected
by the crisis. As a result, directly regressing employment outcomes on the increase in spending
following the policy change could lead to endogeneity concerns. To address this issue, I employ
a two-stage least squares regression, and use the policy change as an instrument for local
governments’ bond issuance.

I find that each additional million dollars of spending resulting from the change in federal
policy creates approximately 22 jobs per year. This effect is concentrated in the private sector,
with limited impact on government employment. Within the private sector, the majority of the
job creation occurs in the services sector. This aligns with the notion that when estimating local
multipliers, the benefits to the tradable goods sector may spill over into neighboring regions,
resulting in lower and less precise estimates compared to those observed in the non-tradable
sector.

While the main analysis focuses on issuers that were already designated as small qualified,
the policy also affected issuers raising debt in the $10 to $30 million range, as they became
eligible to designate their bonds as bank-qualified. Did these issuers benefit from the policy?
Comparing the changes in aggregate municipal bond issuance between these governments and
the qualified non-bunching issuers, I find that the policy change did not induce these issuers to
raise more debt. This can be attributed to the fact that larger governments were not previously
constrained and had no difficulty in accessing the traditional bond market prior to 2009. As a
result, the policy intervention had limited impact on them.

Did the policy change affect banks’ investment decisions? Anecdotal evidence suggests that
bank-qualified municipal bonds are predominantly placed with local banks. Building on this
observation, I construct a Bartik-like instrument to estimate the causal relationship between
banks’ holdings of municipal bonds and their other portfolio decisions.

I define a measure of banks’ exposure to municipal bonds as the inner product of the share
of a bank’s total deposits in a county and the amount of bank-qualified bonds issued in that
county. This measure reflects the average dollar amount of municipal bonds available to a bank,
weighted by the bank’s deposit shares across counties. I scale this measure by the banks’ total
assets, to obtain a standardized metric of a bank’s intensity of exposure per dollar of assets.
To ensure that the variation is not driven by concurrent changes in the deposit share, I fix the

3
shares to the year before the policy change.

To isolate the effect of increased public debt holdings on the supply of credit, I focus on small
business lending under the Community Reinvestment Act (CRA) and mortgage loans reported
under the Home Mortgage Disclosure Act (HMDA). Because banks must report business and
mortgage loans by geography, this allows me to account for demand factors by comparing loans
extended by different banks within the same county. While the impact on small business lending
is muted, I find that banks reduce their mortgage origination activity as a result of an increase
in holdings of municipal bonds.

Overall, the findings in this paper suggest that the policy was effective in supporting small
local economies during the recession. Each additional $1 million in bank-qualified financing
generated approximately 22 jobs across all sectors, yielding a job multiplier of 22. When apply-
ing this multiplier to the aggregate $3.16 billion in bank-qualified debt issued nationwide due to
the policy,2 this resulted in an estimated 70,000 jobs created or saved due to the policy change.
At the county level, this translates to an average of about 252 jobs created per county. The
implied cost to a local government to create a job per year was around $45,500 ($1 million/22).

The job creation estimates align with the wide range of local fiscal multipliers reported in the
existing literature, which typically vary between 7.6 and 39 jobs per million dollars of spending
(Chodorow-Reich 2019). However, this policy’s aggregate impact is relatively modest compared
to larger federal programs. This is not surprising considering the policy only benefited smaller
towns and districts.

This paper contributes to the literature on federal government interventions during economic
recessions. A substantial body of research has examined the impact of the American Recovery
and Reinvestment Act (ARRA) on employment (see eg. Chodorow-Reich et al. 2012; Dube
et al. 2014; Dupor and McCrory 2018; Wilson 2012). Beyond studies on ARRA, extensive
research has investigated whether federal spending can stimulate the economy (e.g., Ramey
2011; Leduc and Wilson 2012; Nakamura and Steinsson 2014). However, by focusing specifically
on transfers or federal government purchases, relatively less attention has been paid to federal
measures designed to alleviate borrowing constraints of local governments. This paper extends
the literature by analyzing a policy aimed at expanding access to financing for small local
governments during economic downturns. While the overall impact of this policy on total
employment is substantially smaller compared to large-scale federal stimulus interventions like
2
This figure is calculated as the total amount of bank-qualified debt issued by bunching local governments
in excess of the $10 million threshold.

4
ARRA, its associated job multiplier — measured as jobs per million dollars of spending — is
comparable to those estimated for federal transfer programs.

This paper also contributes to a growing literature on municipalities’ financial constraints

and investor segmentation. Babina et al. (2020) explore the tax-induced home-state bias in
retail investors’ municipal bond investments and its impact on bond prices. Bergstresser and
Orr (2014) document an increase in commercial banks’ investment in municipal bonds. Yi
(2021) examines how shocks to the supply of municipal credit affect the quality of local public
goods. Adelino et al. (2017) investigate the effects of a municipal bond ratings recalibration on
local spending and employment. This paper adds to these studies along three dimensions: it
documents the presence of segmentation among bank and non-bank investors in the municipal
bond market, it evaluates the effectiveness of bank-financed local government spending, and it
quantifies the associated crowding-out effects.

This article also contributes to the literature examining the relationship between sovereign
debt and bank lending. One strand of this literature has explored the link between sovereign
risk and bank bailouts (Acharya et al. 2014). Other studies have demonstrated that exposure
to distressed sovereign debt can adversely affect lending to the private sector (e.g., Altavilla et
al. 2017) and that increased bank holdings of public debt may displace private lending (Broner
et al. 2014; Becker and Ivashina 2018; Ongena et al. 2019; Pinardon-Touati 2021). This paper
adds to the discourse by investigating whether incentivizing banks to hold municipal bonds
leads to a crowding-out of private lending.

Finally, this paper contributes to the literature on the effects of tax policy in the municipal
bond market. Prior studies have primarily focused on the impact of federal and state tax
exemptions on asset returns and risk-sharing (e.g., Babina et al. 2020; Green 1993; Trzcinka
1982; Green et al. 2007; Ang et al. 2010). In contrast, this paper examines a discontinuity in
the tax treatment of banks’ municipal bond holdings and analyzes the implications of a change
in the tax code for local employment and lending.

5
 II. Institutional Setting

A. Tax Treatment of Municipal Bonds for Banks

Municipal bonds are used by state and local governments to finance public services such as
infrastructure, education, healthcare, and public safety, with the total municipal debt out-
standing amounting to approximately $4.1 trillion. More than 93% of these bonds are exempt
from federal taxation, reflecting the principle of state sovereignty and the separation of powers
codified by the Revenue Act of 1913.

However, the tax exemption does not apply equally to all types of investors. Banks receive
the full tax exemption only when they purchase bonds from bank-qualified municipal issuers.
To obtain the bank-qualified status, a municipal issuer must not raise more than $10 million
in a calendar year. If the total issuance exceeds this $10 million threshold, the issuer loses its
bank-qualified status, making all bonds issued within that year non-qualified.

When banks purchase bank-qualified municipal bonds, they benefit from full tax advantages:
the coupon payments received are tax-exempt, and the bank can deduct (most of) the interest
expenses incurred to acquire the bonds. In contrast, purchasing bonds from a non-qualified
issuer results in banks losing the interest expense deduction (also known as the pro-rata dis-
allowance). This differential tax treatment is specific to banks. Households and non-bank
institutional investors receive full tax exemptions regardless of the issuer’s qualification status.
Sections A.I and A.II in the Appendix provide a detailed description of the bank qualification
and the relevant tax code for banks in the context of municipal bonds.

B. Bank Qualification and the American Recovery and Reinvestment Act

In February 2009, Congress temporarily raised the limit for bank-qualified municipal bonds
from $10 million to $30 million, with this threshold reverting to $10 million after December 31,
2010. The goal of this policy was to enhance local governments’ access to financing by making
it more attractive for banks to purchase municipal bonds, thus supporting municipalities in
maintaining public services and infrastructure projects during the economic downturn.

The change in the bank qualification policy came as part of the broader American Recovery
and Reinvestment Act (ARRA), aimed at supporting the economy during the Great Recession.
As a comprehensive stimulus package, ARRA included a vast array of measures, among which

6
tax reductions and transfer payments to individuals and businesses, as well as substantial
financial assistance to state and local governments, with the majority of the state and local
fiscal relief going to Medicaid and education spending (Congressional Budget Office 2015).3

III. Data

Municipal bond issuance data comes from Ipreo MuniIC. The MuniIC platform covers municipal
bonds issued since the year 2000. The dataset contains information on the issuer, issue and
bond-level size, the offering type and type of bid, the sale date, dated date and maturity
date, as well as coupon value and coupon frequency, yield, and tic details, ratings from S&P,
Moody and Fitch, information on the tax status of the bond and its bank-qualification, the
full redemption call description (first and last call date, and type of call price, e.g. at par),
refunding information, the use of funds description as extracted from the issue prospectus,
details on the presence of insurance or credit enhancements, names and details of the obligor,
financial advisor, bond counsel and paying agent, and finally details on the type of bond (e.g.
general obligation, revenue, BAB, bank-qualified).

Employment data comes from the BLS QCEW; this is census data, it is collected under the
Unemployment Insurance (UI) programs of the United States, and represents around 99.7% of
civilian employment in the country. Population data comes from Census. House price data
comes from the Federal Housing Finance Agency.

Data on banks’ holdings and income statements comes from Call Reports. Data on deposits
comes from the Summary of Deposits (SOD) from the FDIC. Data on mortgage lending comes
from the CFPB and is collected under the Home Mortgage Disclosure Act (HMDA), which
requires financial institutions to disclose loan-level information on mortgage activity. Data
on small business lending comes from the FFIEC, and is collected under the Community and
Reinvestment Act (CRA). CRA loans are defined as loans equal or smaller than $1 million, and
they are either commercial or industrial loans, or loans secured by non-farm, non-residential real
estate. In 2005, all banks with more than $1 billion in assets had to report their CRA lending
activity. Financial institutions have to report their lending activity by geography. Lending is
reported as a flow, not as a stock. Private lending data in this paper corresponds to the amount
of new loans extended by each bank in each county where it operates.
3
Section V.C discusses how to interpret the findings in this paper within the context of the broader federal
intervention.

7
 IV. Bank Qualification and Municipal Issuance

In this section, I present aggregate evidence showing that banks respond to tax incentives in the
municipal bond market, which in turn has significant implications for municipalities’ borrowing
costs and issuance decisions. I then quantify these effects using a bunching estimation approach.

A. Aggregate Evidence

Data from Call Reports on banks’ municipal bond holdings shows that banks respond to the
tax discontinuity. Figure 1 reports banks’ holdings of municipal bonds, expressed as a fraction
of total assets. Consistent with the tax incentives, banks primarily hold bank-qualified bonds,
which comprise between 4% and 5% of their total assets. In contrast, non-qualified bonds make
up just over 0.5% of banks’ assets on average.4

Figure 2a plots the distribution of issuers from 2000 to 2008, with the x-axis representing
the size of debt issuance per year (in $500k increments) and the y-axis showing the number
of municipal issuers in each size bin. The figure reveals a significant spike in the number
of issuers just below the bank qualification limit (bunching), accompanied by a noticeable
absence of issuers immediately above the $10 million threshold.5 Notably, Figure 2b plots the
same distribution of issuers for the years 2009-2010, when the bank qualification threshold was
moved to $30 million, showing a remarkable drop in the mass of issuers around the $10 million
mark. These figures suggest that that the federal tax policy limits on bank qualification affect
municipal behavior. Local governments adjust their issuance sizes to stay within the limit,
thereby retaining access to bank investors, rather than issuing larger amounts and losing access
to banks.

Is it cheaper to borrow from banks? Figure 3 compares the spreads across different issuance
sizes around the bank qualification cut-off for general obligation bonds prior to 2008. Given
the federal tax exemption, I tax-adjust the yields on municipal bonds to obtain federal taxable
equivalent municipal yields and calculate the spreads. Spreads are calculated as the tax-adjusted
4
This analysis excludes municipal loans. Although the importance of the loan market has increased since the
financial crisis (Ivanov and Zimmermann, 2019), municipal loans remained a relatively small portion of banks’
assets, similar to non-qualified bonds, up to 2010. Section A.III in the Appendix provides a detailed description
of how to calculate the breakdown of municipal bond holdings from Call Reports.
5
The issuance distribution also shows smaller spikes at round numbers. These are not associated with a
regulatory discontinuity but rather reflect a common tendency among issuers to prefer round numbers. The
bunching at the $10 million threshold, however, is more pronounced than the rounding behavior. The bunching
estimation in section IV.B accounts for and isolates this rounding behavior.

8
yields over maturity-matched treasuries.The x-axis represents the size of the debt issuance
(in $1 million increments), and the coefficients are expressed as deviations from the average
spread of the $10 million issuances after controlling for bond characteristics including size,
maturity, insurance status, ratings, use of proceeds, as well as year and issuer fixed effects.
Standard errors are clustered by year, and 90% confidence intervals are shown. There is a
significant increase in spreads beyond the bank qualification limit, with non-qualified bonds
having spreads approximately 10 basis points higher than bank-qualified bonds. Considering
the average spread is 132.7 basis points, this reflects an increase of roughly 7.5%. This finding
indicates that borrowing costs are lower in the segment of the market dominated by banks.

Taken together, this evidence suggests that the tax discontinuity leads to investor segmen-
tation in the municipal bond market, which in turn influences interest rates and the financing
decisions of local governments.

B. Bunching Estimation

In order to guide intuition, Section A.IV in the appendix presents a stylized one period
model of local debt financing that provides a conceptual illustration of the mechanisms that
give rise to the bunching phenomenon just shown. The model shows that the banks’ taxation
discontinuity affects equilibrium interest rates on municipal debt, which in turn creates a notch
in municipalities’ budget constraints at the bank qualification limit. When faced with the notch,
affected governments downsize their debt issuance and bunch at the regulatory limit.

In what follows, I detail the bunching methodology used to estimate the fraction of munici-
palities impacted by the tax-induced limits on bank financing and to quantify the corresponding
adjustments to the debt issuance.

Focusing on the pool of municipal bonds issued during the years 2000 to 2008, I express
issuance size (per calendar year) in logs and center the distribution around the 10M limit (in
logs), B ∗ .6 I group the normalized bond issuances in buckets centered at values bj , where
j = −J, ..L, ..0, .., U, ..J, and L and U index the limits of the excluded region around the notch.
Defining nj as the number of municipalities per bin, the estimation follows:
6
I focus on tax-exempt general obligation bonds since there are regulatory and legal limits to the type of
revenue bonds that are allowed to receive the bank qualification status.

9
 p U
γk 1{bk = bj } + η 1{r ∈ R} + ej
X X X
nj = βi (bj )i + (1)
i=0 k=L r∈R

The term bj represents the average distance (in logs) within bucket j between the bond
issuance size in bin j and the cut-off limit for bank-qualification. The first term in the regression
is a p-order polynomial that fits the observed distribution in the data. The second term instead
excludes the region [bL , bU ] around the notch, which is distorted by the bunching behavioral
response. Finally, the third term fits fixed effects for a set of bond issuance sizes.7

The estimate of the counterfactual distribution is hence defined as the predicted bin counts
n̂j omitting the contribution of the dummies in the excluded region, but including the contri-
bution of the round-number fixed effects:

p
η̂ 1{r ∈ R}
X X
n̂j = β̂i (bj )i + (2)
i=0 r∈R

Excess bunching due to the bank-qualification notch is estimated as the difference between
the observed and the counterfactual bin counts within the excluded range to the left of the
cut-off:

0
X 0
X
D̂ = (nj − n̂j ) = γ̂j (3)
j=L j=L

It is possible to define an estimate of missing mass to the right of the limit as

U
X U
X
M̂ = (n̂j − nj ) = − γ̂j . (4)
j>0 j>0

The estimated excess and missing masses, D̂ and M̂ , need not be identical: the policy
might have had both intensive and extensive margin effects, that is it might have induced some
municipalities to under-issue (intensive margin), but it might also have pushed some out of the
market, preventing them to borrow (extensive margin). The estimate of the excess bunching,
7
Municipalities tend to issue in round numbers. The rounding is evident at multiples of 5M, which are then
constitute the set R. The bank-qualification threshold ($10M) falls within the set R of multiples. This implies
that estimating the counterfactual density without controlling for rounding, would overstate the behavioral
response at the notch. The latter term in the specification hence serves the purpose of disentangling the
behavioral response from the round-number bunching. This is possible since the other round numbers, r ∈ R,
are not points of salience for regulatory purposes; in other words, they do not constitute a notch.

10
D̂, provides the intensive margin response, in terms of the number of resized bonds, while the
extensive margin effects are captured by the difference M̂ − D̂.

The core quantity of interest is then ∆B ∗ , that is the behavioral response of the marginal
bunching municipality measured as the percent reduction in the municipal bond size given the
bank-qualification policy limit. This is calculated as:

D̂
∆B ∗ = (5)
ĥ0 (B ∗ )

P0
with ĥ0 (B ∗ ) = j=L n̂j /| b0 −b
L
L
| being the counterfactual density of municipality-bond pair
in the bunching region.

I calculate standard errors using the bootstrap procedure presented in Chetty et al (2011):
I draw with replacement from the estimated errors from equation (1) and generate a new
set of bin counts, which I use to re-estimate the bunching, and proceed by iteration. The
standard errors are estimated as the standard deviation of the estimated parameter in the k-
iterations. I set k to 10,000. The preferred specification uses a 13-degree polynomial. I set
the bin width to 5%, corresponding to $500k steps. Finally, the estimation requires to specify
the limit of the exclusion region. I choose the limits to minimize the difference between the
bunching mass and the missing mass, in line with Kleven and Waseem (2013). This is akin to
estimating a specification where extensive margin responses are minimized. I consider this to
be a reasonable specification, given that the vast majority of municipalities consistently enjoy
credit ratings higher than A-, making it unlikely for an issuer to be unable to find a non-bank
investor and be driven out of the market altogether. Specifically, I estimate (1) on a grid of
all possible combinations of L and U , respectively in [−J, ..j.., 0) and (0, .., j, .., J]; the limits of
the excluded region are such that |M̂ − D̂| is minimized. In robustness specifications, I allow
for the possibility of extensive margin effects.

C. Bunching Results

In 2009-2010, the bank-qualification limit was raised from $10 million to $30 million. This
allows me to assess the validity of the counterfactual density recovered through the bunching
estimation, that is the issuance distribution had the $10 million cut off not been present.
Figure 4 compares the standardized distribution of issuers for the period 2009-2010 against the
counterfactual distribution estimated using the bunching technique. The counterfactual density

11
closely tracks the actual distribution of issuers for the period after the policy change, providing
strong evidence in support of the estimation in Section IV.B.

Figure 5 shows the results of the bunching estimation. This figures plots the empirical and
estimated counterfactual distributions, covering the period 2000 to 2008. The x-axis reports
the size of the municipal bond issuance, while the y-axis reports the number of issuers in
each bin. Each bin represents a 5% incremental deviation from the cut-off, corresponding to
$500k steps. The dashed vertical lines indicate the region affected by bunching. The observed
distribution exhibits non-smooth mass at multiples of $5 million, in line with round-number
issuance. The fitted polynomial appears to do a good job of capturing no-notches spikes in the
distribution. Bunching is especially sharp, even after accounting for round-number issuance.
There is considerable excess mass to the left of the cutoff, and missing mass to the right of the
threshold.

Table I presents the estimation results, along with several robustness tests. In addition to
significance levels and standard errors (shown in parentheses), the table provides 95% confidence
intervals for each estimated parameter.8 Column (1) displays the results from the preferred
specification. The estimated behavioral response, ∆B ∗ , indicates that the average marginal
bunching issuer reduces the size of its municipal bond issuance in the year by 2.93%. Translating
this behavioral response into an intensive margin estimate, D̂/N̂ + , suggests that approximately
41% of issuers—expressed as a fraction of the missing region— who would have issued bonds
exceeding $10 million chose to adjust their issuance to meet or fall below the qualification
limit. Both estimates are statistically significant at the one-percent level. The upper limit of
the exclusion region, $13.5 million, provides the upper bound on the behavioral response of the
affected issuers, indicating that the largest impacted issuer would have issued a bond roughly
28% larger in the absence of the bank qualification limit.9

Columns (2)-(4) report results under alternative specifications that vary the boundaries of
the exclusion region. Across different models, extensive margin effects are both economically
and statistically insignificant, implying municipalities are not driven out of the market due to
the bank qualification limit. Moreover, the behavioral response remains stable in the range of
about 2% to 3%, and significant, indicating local governments chose to reduce the capital they
raise in a year to maintain the bank qualification status.
8
The confidence intervals are calculated using the 250th highest and lowest realizations from 10,000 bootstrap
iterations.
9
This quantity corresponds to the variable bU , as defined in Section IV.B, which is calculated as the distance
(in logs) between the mid-point of the ($13 mil, $13.5 mil) upper bucket and the $10 million cut-off.

12
D. Borrowing Dynamics

The bunching estimation indicates that local governments respond to tax-induced limits on
bank financing by reducing their bond issuance in a year to meet the bank qualification limit.
This behavior raises the question of whether bunching is part of a dynamic strategy. While the
reduction in issuance might imply that governments forgo additional funding, it is also possible
that issuers are dynamically adjusting their borrowing, spreading their total financing needs
over multiple years to remain within the $10 million cutoff.

To explore this, I examine the borrowing patterns of bank-qualified issuers that bunch,
and compare them to those that do not bunch. Table II shows that local governments that
bunch issue bonds more frequently than those that do not. On average, bunching governments
issue bank-qualified bonds 2.8 times over the 2000-2008 period, compared to 1.8 times for
non-bunching governments. This leads to a shorter interval between issuances for bunching
governments, with an average of approximately 1.8 years as compared to 2 years for non-
bunching issuers, assuming both groups issue more than once.

The higher frequency of issuance among bunching issuers may not necessarily reflect a
deliberate strategic choice. Bunching issuers are, on average, more than three times larger than
non-bunching issuers in terms of population (Table III Panel A). Thus, the increased frequency
of bond issuance could simply be a response to the greater financing needs of these larger
issuers.

In addition, while bunching governments do raise funds more frequently, most issuers do
not access the bond market in consecutive years. Figure 6 shows the distribution of issuance
for bunching (Panel A) and non-bunching (Panel B) issuers, conditional on having issued the
year before. Among those issuing bank-qualified bonds and bunching the year before, less than
10% is bunching again in the following year, while around 67.5% do not issue any bonds in
the subsequent year. This suggests that strategically spreading bond issuances over consecutive
years is not a prevalent strategy among bunching issuers. In comparison, 75.8% of non-bunching
issuers also do not issue bonds in the year following an issuance.

Overall, it is still possible that for a small set of issuers the bunching behavior reflects a
dynamic strategy, involving delayed issuance rather than solely under-issuance. It is important
to note that local governments incur a variety of costs and fees when issuing bonds, including
payments to municipal and legal advisors, trustees, agents, auditors, rating agencies, and other
service providers. These costs typically average around 1.7% of the face value of issuance, with

13
smaller offerings facing significantly higher costs, potentially up to 8.5% (Joffe 2015). Therefore,
regardless of the strategy employed, the bank qualification limit imposes costs on issuers. These
costs arise either from the need for multiple issuances, which delays real investment and increases
fees and associated expenses, or from the restriction on accessing additional funding.

V. Real Effects of the Bank Qualification Expansion

In this section, I investigate the impact of the federal expansion of tax benefits for banks
investing in municipal bonds. I provide evidence that the change in the bank-qualification policy
led to increased municipal borrowing and estimate its corresponding effects on employment.

A. Empirical Design

The identification strategy exploits the cross-sectional heterogeneity in municipalities’ ex-

posure to the change in the bank qualification limit. Figure 7 illustrates the empirical setting.
I focus on bank-qualified issuers, comparing municipalities that were bunching before the reg-
ulatory change (region (2) in the figure) with non-bunching qualified issuers (region (1)).10
Following the estimation in Section IV, I define issuers in region (2)—the bunching issuers—as
the local governments impacted by the regulatory shock. For these governments, the policy
change effectively relaxed a borrowing constraint. I define municipalities in region (1) as con-
trols because they also rely on bank borrowing, allowing the investor type to remain constant,
but their borrowing levels were consistently below the bunching region, suggesting the bank
qualification policy did not bind for these issuers. The analysis covers the period from 2004 to
2010.

To estimate the effects on employment and wages, I aggregate municipalities at the county
level.11 I define Intensity as the fraction of affected municipalities in a county. Treated counties
10
The bunching region is derived from Table I, column (1). Results remain consistent when varying the
sampling restrictions and the limits of the affected region. These results are presented in Table A3 in the
Appendix. Specifically, columns (1), (4), and (7) remove the sampling restriction to include also non-urban
counties. Columns (2), (5), and (8) exclude a buffer region of issuance between treated and control groups.
Finally, columns (3), (6), and (9) vary the bunching threshold according to Table I column (4).
11
The aggregation at the county level is needed because an extra dollar of spending will have spillover effects
in neighboring areas, with the effects being larger the smaller the municipality is. A municipality can be thought
of as a small open economy, therefore the impact of an extra dollar of spending will be affected by expenditure
switching as well as by migration forces. Forcing the impact of government spending to be circumscribed within
the boundaries of a town is therefore likely to result in uninformative estimates. For these reasons, I also focus

14
are then the counties with at least one bunching issuer, Intensity > 0. Control counties are
those where Intensity takes value zero. To correctly locate where the spending is occurring, I
focus on single-county issuers, and exclude authorities and districts that raise funds for multiple
geographical areas. The final sample includes counties where at least one government has issued
bonds at any given year. Table A1 in the appendix details how the sample changes with the
sampling restrictions.

I estimate a 2SLS, where the first stage is the impact of the federal policy change on local
governments’ issuance, and the second stage is the effect of a marginal dollar of (instrumented)
bank qualified debt on local employment:

Issuancei,t = b1 Intensityi x P ostt + b2 Xi,t + ai + asize,t + ei,t (6)

Yi,t = β1 Issuance
\ i,t + β2 Xi,t + ai + asize,t + ξi,t (7)

Issuance is the sum of bank qualified debt raised by region (1) and region (2) municipalities
within each county. The variable P ost takes value of one during the regulatory change (2009-
2010), and zero otherwise.12 The variable ai represents county fixed effects.

A concern may be that bunching and non-bunching municipalities could have a differential
exposure to the recession, which could drive the observed employment effects. Including state-
by-year fixed effects could potentially alleviate this concern. However, the sample studied
consists of 35 states (as shown in Table A2 in the appendix), with 16 states having fewer than
five treated or control counties, and 14 states containing only treated or only control counties.
Due to this limited sample size, I am unable to include state-by-year fixed effects, and instead
include population decile-by-year fixed effects, asize,t , to account for time-varying factors that
could differentially affect highly versus scarcely populated counties.

The dummies Intensity and P ost are subsumed by respectively the county fixed effects and
the size-by-year fixed effects. In the specifications where I do not include the size decile-by-year
f.e., P ost is not absorbed and will be present in the regressions.

Since local governments rely on local property taxes to repay bonds, to further account for
time varying local economic conditions, I include a set of extra controls, Xi,t , namely the house
price index (HPI) and county population. The instrument for bank-financed debt issuance is
on the subset of urban counties defined as having a population of at least 25,000.
12
The law change was first proposed in January 2009 and officially passed in February 2009.

15
given by Intensityi x P ostt . Yi,t represents the log of employment and the log of total wages in
a county. Standard errors are clustered at the county level.

B. Univariate Analysis

I start by investigating the presence of pre-existing differential economic trajectories between

treatment and control counties, as well as differences across local governments’ budgets. To do
so, I hand match issuers to the Census of states and local governments. All local governments
are present during census years, while only a small subset is sampled for the survey on non-
census years. Smaller municipalities are sampled at a much lower frequency. For this reason,
when inspecting government finances, I focus on the most recent census year prior to the policy
shock, which is 2007.

Table III Panel A reports budget characteristics for the sample of municipal issuers hand-
matched to Census. Issuers are classified into bunching (region (1)) and non-bunching (region
(2)), following Section V.A. Both are bank qualified issuers. Bunching issuers are significantly
larger than non-bunching ones, which is not surprising given the size of their debt issuance.
However, the operating budgets of the two set of issuers appear remarkably similar when looking
at per capita revenues, expenditures and debt outstanding.

Table III Panel B reports pre-treatment trends for the set of treated and control counties.
The differences are estimated both with and without population deciles by year fixed effects.
The two sets of counties appear to be on similar economic paths before 2009 in terms of
overall employment and wage growth. House price growth also appear both statistically and
economically comparable, after controlling for size decile by year fixed effects. Taken together,
Table III shows a similar fiscal profile for municipalities and comparable county-wide economic
patterns, providing support for the validity of the estimation.

Table IV reports summary statistics for the main variables in the analysis. On average,
approximately 13.6% of issuers in a county are bunching before the regulatory change. This
corresponds to an average (median) of roughly 3 (2) issuers per county. Bunching and non-
bunching issuers raise an average of $23.5 million in bank qualified debt in aggregate per
county. Within each county, the private sector makes up for most of the employment, while
government employment constitutes only a small fraction of the total. Within the private
sector, employment in the service industry is more than four times that in the goods sector, a
pattern also reflected in the total wage bill.

16
C. Results

Table V reports the results from the first stage regression, layering controls and a more stringent
fixed effect structure. Issuance is expressed in hundred million dollars. The impact of the federal
policy change on municipal issuance is both economically and statistically significant. The
coefficient in column (3) indicates that the policy change generates an increase in bank qualified
issuance of $2.7M per one standard deviation (0.121) increase in the fraction of bunching
municipalities in a county (Intensity).

Table VI reports the estimation results for employment. The dependent variables are ex-
pressed in logs, so the estimated coefficient is an elasticity. Issuance is the instrumented bank
qualified debt. The coefficients are stable across specifications. Under the most stringent set
of controls, the coefficient in column (3) indicates that each marginal $1M of bank qualified
financing generates (or saves) around 22 jobs across all sectors. The effect on employment is
obtained by multiplying the estimated elasticity (0.193) by the average ratio of employment to
government debt (112.42). The employment effect is concentrated in the private sector - just
under 21 jobs (0.219*93.17) from column (6) - while there is limited impact on government
employment.

Analyzing the use of proceeds gives an insights into how the funds were spent, which can help
interpret the employment outcomes. Table A4 in the appendix presents the breakdown of the
use of proceeds from bank-qualified bonds issued in 2009-2010. The table shows that over 41%
of bonds were allocated to general purposes, approximately 33% were directed toward primary
and secondary education infrastructure, and 13% were used for water and sewer projects. The
remaining funds were distributed across various projects, including toll roads, sanitation, parks
and zoos, utilities, and airports.

To further analyze the employment effects, I decompose the local employment response into
the goods and services sectors (Table VII). The results indicate that job creation is exclusively
concentrated in the non-tradable sector, with no significant impact observed in the goods sector.
This finding aligns with the typical use of proceeds of municipal bonds. Moreover, when
estimating local multipliers, the benefits to the tradable goods sector may spill over into other
regions, leading to lower and less precise estimates compared to the non-tradable sector.

I then investigate whether the policy change had an immediate or delayed impact on is-
suance and employment. Figures 8 and 9 display the effects on municipal bond issuance and
employment using a calendar-time coefficient plot. These plots show the regression coefficients

17
of the logarithm of the main variable of interest on the interaction between treated intensity
and year dummies, covering the period from 2004 to 2010. The year 2008 serves as the base
year, and the results are presented with a 90% confidence interval. The effects on employment
are disaggregated into government, private, and service sectors. As shown in the figures, the
effects on employment become more pronounced by 2010. This delayed response is consistent
with the nature of public financing, which often involves a lengthy process due to the need for
municipalities to comply with public bidding and procurement regulations, potentially delaying
the start of infrastructure projects.

Table A5 in the Appendix reports the impact of bank qualified issuance on total wages.
Similarly to the employment estimates, the effect on wages is obtained by multiplying the
estimated elasticity by the average ratio of wages to government debt. The results imply a
wage bill multiplier of 0.70 for both the private sector and across all sectors, however these
effects are imprecisely estimated.

To further address concerns that results may be confounded by counties experiencing the
housing crash to different extents, in addition to including size by year fixed effects, I exclude
from the analysis the states most affected by the foreclosure crisis. Based on 2009 data from
RealtyTrac, the states with the highest percentages of properties receiving foreclosure filings
were Nevada, Arizona, Florida, California, and Utah. Given that Nevada and Florida are not
included in my sample, I exclude the remaining states (Arizona, California, and Utah) from
the analysis. The findings of the paper remain robust even after excluding these states (Table
VIII).

In addition, I show that there are no significant differences in the growth of the house price
index (HPI) between treated and control counties. Figure 10 presents the results, and reports
the regression coefficients of the logarithm of the HPI on the interaction between treated and
year dummies, covering the period from 2004 to 2010. This regression controls for population,
county fixed effects, and size decile-by-year fixed effects, with 2004 as the base year.

Overall, the evidence suggests that treated and control counties were not significantly dif-
ferent in their exposure to the 2007-2009 financial crisis. This finding may be attributable to
the fact that the foreclosure crisis disproportionately impacted densely populated regions (Pew
Research Center 2009), while the issuers in this study are predominantly small to medium-sized
jurisdictions.

The analysis thus far has focused on local governments that were already designated as

18
qualified small issuers prior to the policy change (zones 1 and 2 in Figure 7). However, the
policy also impacted issuers raising funds in the $10 to $30 million range, as they became
eligible to designate their bonds as bank-qualified (zone 3 issuers). Did these issuers benefit
from the policy? To explore this, I estimate Equation 6 for zone 3 issuers, defining Intensity
as the fraction of zone 3 issuers in a county. The same set of controls is applied, namely the
small qualified issuers in zone 1. I then compare the changes in aggregate municipal bond
issuance between treated and control counties before and after the policy change. The results
are presented in Table A6 in the Appendix. The findings show that the policy change did
not result in increased borrowing for zone 3 issuers. This can be attributed to the fact that
zone 3 issuers were not previously constrained and, as larger governments, they already had no
difficulty accessing the traditional bond market. As a result, the policy intervention may have
had limited impact on them.

When interpreting the estimates, it is important to consider that the bank qualification pol-
icy was implemented as part of the larger American Recovery and Reinvestment Act (ARRA).
There is a possibility that state and local governments reduced their own spending and replaced
it with federal funds and grants. This substitution effect suggests that the observed increase in
bond issuance may represent a lower bound for the policy’s impact in the absence of concur-
rent federal transfers. Conversely, the simultaneous increase in federal spending through funds
disbursed to state and local governments could complicate the interpretation of the employ-
ment results, particularly if such federal spending is correlated with local government spending
decisions.

However, it is important to note that the disbursement and expenditure of ARRA funds
were notably slow. While Medicaid funding was among the quickest to be utilized (Chodorow-
Reich 2012), the Department of Transportation (DOT) spending - which is highly relevant to
local government expenditures - was among the slowest to be disbursed (Congressional Research
Service 2020). The majority of DOT funding was expended in 2010 and 2011. This slow pace of
disbursement provides some reassurance that the employment effects observed are not primarily
driven by ARRA-related spending.

Finally, to the extent that treated and control counties in the sample were likely not differ-
entially exposed to the crisis, it is possible that the growth in federal transfers they received
did not significantly differ, reducing concerns of possible confounders.

Taken together, the results in this section indicate that the policy change was effective in
stimulating local governments’ issuance, with consequent positive effects on local employment

19
growth.

VI. Impact of the Federal Policy on Private Lending

Federal policies that incentivize banks to hold government debt may have the unintended
consequence of crowding out private lending, which in turn may reduce the effectiveness of
the policy in stimulating the economy (see eg. Broner et al. 2014; Becker and Ivashina 2018;
Ongena et al. 2019; Huang et al. 2019). In this section, I examine whether the increase in
municipal bond holdings by banks led to a reduction in credit extended to private firms and
households.

A. Empirical Design

To examine whether an increase in municipal debt holdings leads to a decrease in private

lending, I exploit the heterogeneity in banks’ exposure to bank-qualified issuance. Anecdotal
evidence suggests that municipalities are more likely to place their bonds with banks that have
a local presence rather than with banks that have no local operations. Based on this anecdotal
evidence, I define the following Bartik-like measure:

P
j DepositSharei,j ∗ BQ Issuancej,t
Bank Intensityi,t = (8)
Assetsi
BQ Issuance represents the total dollar amount of bank-qualified debt issued in county j
each year t. DepositSharei,j is the fraction of bank i’s total deposits in county j. To ensure
that the observed variation is not driven by concurrent changes in deposit shares or variations
in banks’ deposit strategies, I fix the deposit shares to their values from the year prior to the
policy change. The numerator represents the average dollar amount of a bank’s exposure to
bank-qualified municipal bond issuance, weighted by the bank’s cross-county deposit shares.
To account for variation in bank size, I scale this value by each bank’s assets, fixed as of the
year prior to the policy change. This provides a standardized measure of exposure to municipal
bond issuance per dollar of assets, Bank Intensity.

Under this measure, banks with a stronger local presence in counties that experience more
pronounced increases in qualified bond issuance, will correspondingly increase their municipal
bond holdings more significantly. To test this, I estimate the following specification:

20
 Log(M uniBondsi,t ) = βBank Intensityi,t + γXi,t−1 + ai + ei,t (9)

Log(M uniBondsi,t ) is the log of one plus the amount of municipal bonds held on bank
i balance sheet, as reported in the December call reports. I aggregate holdings at the bank
holding company level. The term ai denotes bank fixed effects.

To account for the fact that bank branch locations are not random and the possibility that
banks operating in different geographic regions may differ in aspects other than exposure to
municipal bond issuance, I adopt the approach of Chakraborty et al. (2020, 2018). Specifically,
I include the following bank-level characteristics lagged by one year, Xi,t−1 : the equity ratio,
net income scaled by assets, cost of deposits, and cash to assets, all measured as of year t − 1.
These variables are intended to capture time-varying differences in the financial health of banks,
which may influence bank lending behavior. Robust standard errors are clustered at the bank
level.

Next I investigate whether the increase in banks’ holdings of municipal securities impacts
their lending to private firms. As banks might face different demand for loans, to isolate
the effect of increased public debt holdings on the supply of credit, I focus on small business
loans and mortgages. Under the Community Reinvestment Act (CRA) and the Home Mortgage
Disclosure Act (HMDA), banks are required to report their small business and mortgage lending
activities by geographic location. This reporting requirement enables me to control for loan
demand by comparing loans issued by banks operating within the same county.13 I estimate
the following reduced-form specification:

Log(Loansi,j,t ) = βBank Intensityi,t + γXi,t−1 + ai + ηj,t + eijt (10)

Here, Loans represents either small business loans or mortgage loans, which are reported
as flows. Thus, Log(Loansi,j,t ) denotes the logarithm of the total new loans originated by bank
i in county j and time t. I continue to include bank fixed effects, ai , and a vector of balance
sheet characteristics Xi,t−1 , which includes the equity ratio, net income scaled by assets, interest
expense on deposit ratio, and cash-to-asset ratio, all measured as of t − 1. Finally, I include
county by year fixed effects, ηj,t .

This specification enables me to control for loan demand by comparing the lending behavior
13
CRA reporting coverage changed substantially in 2005, so I begin the analysis in that year. HMDA data
spans the years 2006-2010.

21
of two banks — similar on observables of financial health but differentially exposed to bank
qualified bond issuance — within the same county and time period. Standard errors are double-
clustered by county and bank. In the Appendix, I also investigate potential crowding out effects
on Treasury and agency debt.

B. Results

Table A7 in the appendix provides summary statistics for banks with above and below
median values of BankIntensity. Banks with above median BankIntensity are significantly
smaller in terms of overall size of their balance sheet. Figure A1 reports the distribution of
aggregate bank-qualified issuance per county for the years 2009-2010, and shows considerable
variation in the amount of municipal bond issuance across geographies.

Table IX reports the results of the estimation from Eq. 9. The findings indicate that banks
with branches in regions experiencing a higher uptake in bank-qualified issuance, significantly
increased their municipal bond holdings over time. Specifically, a one standard deviation in-
crease in the Bartik-like instrument (1.09) is associated with a 7.7% increase in banks’ municipal
bond holdings, under the most stringent of the specifications.

Is the increase in municipal bond holdings driven by large or smaller local banks? Figure
11 presents the regression coefficients of the logarithm of municipal bond holdings against year
dummies for the period 2005-2010, with 2008 as the base year. The findings are displayed for
different quintiles based on bank asset size. The regression includes bank fixed effects along
with the standard controls from the main specification. The coefficients are shown alongside
90% confidence intervals.

The figure reveals that, following the change in the bank qualification rule, banks of all sizes
increased their municipal bond holdings, whereas their holdings were relatively stable before
the policy change. Notably, small and medium-sized banks increased their municipal bond
holdings more significantly than larger banks. In particular, small banks nearly doubled their
municipal bond holdings starting in 2009.

Table X presents the results of estimating specification Eq. 10. Column (1) displays the
results for CRA small business loans, while columns (2) and (3) show results for mortgages
disaggregated into originated and purchased categories, respectively.

I find no significant evidence of crowding out of small business lending (column (1)). How-

22
ever, the coefficient in column (2) shows that banks reduce mortgage origination in response
to increased municipal debt holdings. Specifically, a one standard deviation increase in expo-
sure to municipal debt (0.38) is associated with about a 7% reduction in mortgage origination.
There is no significant impact of purchased mortgages (column (3)). Table A8 in the Appendix
presents results for Treasury and agency debt holdings. Although the estimated coefficients on
agency debt are negative, indicating a possible reduction in holdings, they are not precisely
estimated.

The identification used in this section exploits banks’ differential exposure to municipal bond
issuance due to their pre-existing deposit shares, and does not require the variation in issuance
itself to be random or exogenous to the initial levels of banks’ asset holdings (Goldsmith-
Pinkham et al. 2020). If unobserved factors drive both municipal bond issuance and loan
demand, the inclusion of county by year fixed effects should absorb them to the extent that
such variation happens at the county level. A potential bias could arise if banks are exposed
to different loan demand in a given county, in a way that correlates with their exposure to
municipal bond issuance.

To provide more insight into the drivers of this shift-share design, Table A9 in the appendix
reports the Rotemberg decomposition of the estimated coefficients, and a number of diagnostics
as suggested by Goldsmith-Pinkham et al. (2020). Reassuringly, the issuance component of the
instrument (gk ) has a non-trivial correlation with the sensitivity-to-misspecification elasticities,
while the variation in deposit shares across banks (V ar(zk )) only explains a small part of the
variation in the sensitivity elasticities (Panel B). Interestingly, Panel C shows that, despite a
very large number of counties, the top five counties in terms of Rotemberg weight account for
about 25% of the total positive weight in the estimator (0.346/1.394).14

Overall, although more muted, the findings in this section echo the evidence in the literature
showing that sovereign debt can crowd out private lending (e.g., Acharya et al. 2014; Bottero
et al. 2017; De Marco 2019).
14
Although a small number of counties is particularly influential for the estimation, there are over 500 unique
banks with deposit branches located in these counties. This extensive bank presence reduces concerns that the
estimation is driven by a limited number of exposed banks.

23
 VII. Overall Impact of the Policy Change

The findings in this paper indicate that each marginal $1 million of bank qualified financing
generated around 22 jobs across all sectors (the job multiplier). The coefficients estimated in
this paper are local effects, but it can be useful to perform a back-of-the-envelope calculation
to approximate the aggregate employment impact of the policy change.

A simple way to estimate the total job creation resulting from this policy is to multiply the
job multiplier estimate by the total spending induced by the policy. The aggregate spending
amounts to approximately $3.16 billion nationwide, and is calculated as the total amount of
bank-qualified debt issued by affected (ie. bunching) local governments in excess of the $10
million threshold. Using the job multiplier estimate of 22 jobs generated per $1 million of
spending, the estimated impact of the policy on aggregate U.S. employment was about 70,000
jobs created or saved (22/1mil times 3.16bils). Analogously, estimating the policy-induced
spending at the county level suggests that approximately 252 (55) jobs were created per county
on average (median), representing roughly 0.26% (0.05%) of total employment in the average
(median) county.15 As for the cost per job created, with 22 jobs generated per $1 million of
spending, the implied cost to local governments per job created per year is around $45,500 ($1
million/22).

How do these findings compare to existing literature on local fiscal multipliers? A substantial
body of research has studied employment multipliers, with estimates ranging from 7.6 to 39
jobs created or saved per million dollars of government spending, yielding an average of 21 jobs
and a median of 19 jobs (Chodorow-Reich, 2019). The job creation estimates in this study fall
squarely within this range. Most of the literature focuses on spending that is either transfer or
windfall financed. For example, Chodorow-Reich et al. (2012) estimate a cost per job of $26,000
based on Medicaid transfers to state governments. Suarez-Serrato and Wingender (2021) derive
a cost per job of $30,000 using variations in federal spending allocations driven by population
estimate errors, while Shoag (2016) estimates $35,000 per job based on windfall state pension
spending.

In contrast, the spending evaluated in this study is deficit-financed, with municipalities

repaying debt through future local tax revenue. It is not surprising then, that the aggregate
spending induced by the policy studied in this paper, and its implied aggregate job creation,
15
When interpreting the county-level estimates, it is important to note that, on average, just over 13% of
issuers in a county are directly impacted by the policy. As a result, the policy’s effect on job creation is expected
to constitute only a small fraction of the county’s total employment.

24
are only a small fraction of the aggregate spending and job creation of other federal programs.
Under this policy, the debt needs to be raised and paid back by the local issuers, which also
implies that it may need to fall within local governments’ budgetary rules and limits. More
importantly, however, the policy mainly affected small towns and districts. Moreover, the
unintended crowding out effects contribute in dampening the effectiveness of the spending.
Few studies have examined the effects of debt-financed spending multipliers. A notable paper
is Adelino et al. (2017), who estimate a job multiplier of 51 per million dollars of municipal
spending, corresponding to a cost per job of approximately $20,000. Although the mechanisms
and point estimates differ between their study and this paper, the wide standard errors in the
analysis suggest that the implied real effects may be of similar magnitude.

Focusing on the policy’s impact on banks, the policy change allowed banks to deduct a
higher fraction of their interest expenses, thus reducing their taxable base. So ideally, one
could do a back of the envelope calculation to estimate banks’ elasticity of municipal bond
holdings to the increase in tax benefits. To estimate this, one could compare the percent
increase in tax deductions with the percent increase in municipal bond holdings. However, this
calculation is not straightforward because the deduction is calculated as a fraction of the interest
expense, which is itself influenced by the federal funds rate. In 2009, the federal funds rate
dropped to almost zero, effectively reducing banks’ overall interest expenses and, consequently,
the actual tax deductions they could claim. As a result, although the deduction rules became
more favorable (an 80% increase)16 , the actual benefit to banks, calculated as the total dollar
value of muni-related tax deductions, decreased due to the lower interest rate environment.17

In the short term, the difference between qualified and non-qualified bonds became smaller
for banks because of the drop in the fed funds rate. However, it is important to note that these
are long-term bonds in a market characterized by a buy-and-hold strategy. Banks are unlikely
to be incentivized to purchase non-qualified municipal bonds based solely on a temporarily low
federal funds rate. This is because these bonds remain on the balance sheet to maturity, and
the associated tax benefits diminish as the federal funds rate rises back.
16
If initially 100% of the interest expense incurred to buy non-qualified bonds was non-deductible, after the
policy change only 20% of the cost is non-deductible. This change represents an 80% difference in the disallowed
portion.
17
As explained in Section A.III, one can back out the dollar value of bank qualified (BQ) and non-qualified
municipal bonds on a bank’s balance sheet. Once the value of BQ bonds is obtained, the deduction can be
calculated as BQBonds
Assets ∗ Int.Exp ∗ 80%, as explained in Section A.II. The average deduction went from $186k
in 2008, to a mean of $137.5k for 2009-2010.

25
 VIII. Conclusion

This paper analyzes the impact of a federal policy that expanded tax benefits for banks investing
in municipal bonds during the 2007-2009 financial crisis. By raising the cap on bank-qualified
bond issuances, the policy aimed to ease borrowing constraints for small local governments.
Unlike direct grants or transfers, this policy provided a tax incentive to banks, while leaving
municipalities responsible for debt repayment.

The findings indicate that the policy successfully increased municipal bond issuance, partic-
ularly for issuers previously constrained by the $10 million limit. This resulted in a significant
increase in job creation, with an estimated 22 jobs generated per million dollars of additional
spending. The majority of these jobs were created in the private sector, primarily in non-
tradable sector.

Echoing previous research on sovereign debt crowding out private lending, this policy had
the unintended consequence of crowding out of banks’ investment in mortgage loans, as a result
of their increase in municipal bond holdings.

This paper contributes to the body of work on federal interventions during recessions, com-
plementing existing studies on direct fiscal transfers. The analysis also adds to the growing
research on municipalities’ financial constraints. Finally, this paper extends the literature on
the relationship between sovereign debt and private lending by examining the municipal bond
market and the effects of tax policy changes on bank investment behavior.

Overall, the findings suggest that tax-based incentives for bank investment in municipal
bonds can effectively support small local economies during recessions.

26
 References

Acharya, Viral, Itamar Drechsler, and Philipp Schnabl, (2014). A Pyrrhic victory? Bank
bailouts and sovereign credit risk. Journal of Finance 69, 2689-2739

Adelino, Manuel, Igor Cunha, Miguel Ferreira (2017) The Economic Effects of Public Financing:
Evidence from Municipal Bond Ratings Recalibration, Review of Financial Studies 30, 3223-
3268

Altavilla, Carlo, Marco Pagano, Saverio Simonelli (2017), Bank Exposures and Sovereign Stress
Transmission, Review of Finance, Volume 21, Issue 6, October 2017, Pages 2103–2139

Ang, A., Bhansali, V., Xing, Y., (2010). Taxes on tax-exempt bonds. Journal of Finance 65,
565-601.

Babina, Tania and Jotikasthira, Chotibhak and Lundblad, Christian T. and Ramadorai, Tarun,
(2020) Heterogeneous Taxes and Limited Risk Sharing: Evidence from Municipal Bonds, The
Review of Financial Studies

Becker, Bo, and Victoria Ivashina (2018) Financial Repression in the European Sovereign Debt
Crisis. Review of Finance 22, no. 1 (February 2018): 83–115.

Bergstresser, Daniel and Peter Orr. (2014) Direct Bank Investment in Municipal Debt. Mu-
nicipal Finance Journal

Bottero, M.; S. Lenzu; and F. Mezzanotti, (2017) Sovereign debt exposure and the bank lending
channel: impact on credit supply and the real economy , working Paper, Bank of Italy

Broner, F., Erce, A., Martin, A., and J. Ventura, (2014). Sovereign debt markets in turbulent
times: Creditor discrimination and crowding-out effects. Journal of Monetary Economics 61,
114-142

Chakraborty, Indraneel and Goldstein, Itay and MacKinlay, Andrew, (2020) Monetary Stimulus
and Bank Lending, Journal of Financial Economics (JFE), Forthcoming

Chakraborty, Indraneel and Goldstein, Itay and MacKinlay, Andrew (2018) Housing Price
Booms and Crowding-Out Effects in Bank Lending, The Review of Financial Studies, Volume
31, Issue 7, July 2018, Pages 2806–2853

Chetty, Raj, John N. Friedman, Tore Olsen, and Luigi Pistaferri, (2011) Adjustment Costs,Firm
Responses, and Micro vs. Macro Labor Supply Elasticities: Evidence from Danish Tax Records,
The Quarterly Journal of Economics, 126 (2), 749-804.

27
Chodorow-Reich, Gabriel; Laura Feiveson, Zachary Liscow, and William Gui Woolston. (2012).
Does State Fiscal Relief During Recessions Increase Employment? Evidence from the American
Recovery and Reinvestment Act, American Economic Journal: Economic Policy, 4 (1), pp. 118-
145.

Chodorow-Reich, Gabriel. (2019). Geographic Cross-Sectional Fiscal Multipliers: What Have

We Learned? American Economic Journal: Economic Policy, 11 (2): 1-34.

Congressional Research Services (2020) Transportation Infrastructure Investment as Economic

Stimulus: Lessons from the American Recovery and Reinvestment Act of 2009

Congressional Budget Office (2015), Estimated Impact of the American Recovery and Rein-
vestment Act on Employment and Economic Output in 2014

De Marco, F. (2019). Bank Lending and the European Sovereign Debt Crisis. Journal of
Financial and Quantitative Analysis, 54(1), 155-182

Dube, Arin, Ethan Kaplan, and Ben Zipperer, (2014) Excess Capacity and Heterogeneity in
the Fiscal Multiplier: Evidence from the Obama Stimulus Package,working paper

Dupor, Bill, and Peter McCrory, (2018), A Cup Runneth Over: Fiscal Policy Spillovers from
the 2009 Recovery Act, Economic Journal, Forthcoming.

Green, R.C., (1993). A simple model of taxable and tax-exempt yield curves. Review of
Financial Studies 6, 233-264.

Green, R.C., Hollifield, B., Schurhoff, N., (2007). Financial intermediation and the costs of
trading in an opaque market. Review of Financial Studies 23, 101-146.

Goldsmith-Pinkham, Paul, Isaac Sorkin, and Henry Swift. 2020. Bartik Instruments: What,
When, Why, and How. American Economic Review, 110 (8): 2586–2624.

Huang, Yi, Pagano, Marco and Panizza, Ugo (2019) Local Crowding Out in China, The Journal
of Finance, 75: 2855-2898

Ivanov, Ivan and Zimmermann, Tom (2019) The ’Privatization’ of Municipal Debt, Working
Paper

Joffe, Marc. Doubly Bound: The Cost of Issuing Bonds. Berkeley, CA: Haas Institute for a
Fair and Inclusive Society, University of California, Berkeley, 2015

Kleven H, Waseem M. (2013). Using notches to uncover optimization frictions and structural
elasticities: theory and evidence from Pakistan. Q. J. Econ. 128:669-723

28
Nakamura, Emi and Jon Steinsson. (2014). Fiscal Stimulus in a Monetary Union: Evidence
from U.S. Regions, American Economic Review, 104 (3), pp. 753-792.

Pew Research Center (2009) Through Boom and Bust: Minorities, Immigrants and Homeown-
ership

Pinardon-Touati Noemi (2021), The Crowding Out Effect of Local Government Debt, working
paper

Ongena, Steven, Alexander Popov, and Neeltje Van Horen. (2019) The Invisible Hand of the
Government: Moral Suasion during the European Sovereign Debt Crisis, American Economic
Journal: Macroeconomics 11, no. 4 : 346–79.

Ramey, Valerie. (2011). Can Government Purchases Stimulate the Economy?, Journal of
Economic Literature, 49 (2), pp. 67385.

Suarez-Serrato, Juan Carlos , and Philippe Wingender. (2021). Estimating Local Fiscal Mul-
tipliers, mimeo, University of California, Berkeley.

Shoag, Daniel. (2016). The Impact of Government Spending Shocks: Evidence on the Multi-
plier from State Pension Plan Returns,mimeo, Harvard University.

Trzcinka, C., (1982). The pricing of tax-exempt bonds and the Miller Hypothesis. Journal of
Finance 37, 907-923.

Yi, Hanyi (2021), Financing Public Goods, working paper

29
 Table I: Bunching at the Bank Qualification Limit
This table reports the results of the bunching estimation from Section IV. Reported in the table
are: the behavioral response (∆B ∗ ), the intensive margin effect (D/N + ), the extensive margin
effect ((M̂ − D̂)/Nˆ+ ), the lower and upper limits of the affected issuance region, and the choice
of polynomial. Bootstrapped standard errors and 95% confidence intervals are in parenthesis.

(1) (2) (3) (4)

Behavioral Response (∆B ∗ ) 0.0293*** 0.0294*** 0.0221*** 0.0324***

st. err. (0.0038) (0.0043) (0.0061) (0.0039)
95% C.I. (0.0221, 0.0375) (0.0215, 0.0389) (0.0117, 0.0365) (0.0253, 0.0409)

Intensive Margin (D̂/N̂ + ) 0.4137*** 0.3447*** 0.2397*** 0.2963***

st. err. (0.0634) (0.0648) (0.1333) (0.0392)
95% C.I. (0.3017, 0.5564) (0.2362, 0.4983) (0.106, 0.5509) (0.2262, 0.3828)

Extensive Margin (M̂ − D̂)/Nˆ+ ) -0.1758 -0.1487 -0.1444 0.0169

st. err. (0.1195) (0.1425) (0.3885) (0.0832)
95% C.I. (-0.4418, 0.0288) (-0.488, 0.0857) (-1.0657, 0.2398) (-0.1646, 0.1626)

Exclusion Limits ($M) (8.5, 13.5) (8.5, 14.5) (8, 17.5) (9, 13.5)

Polynomial 13 13 13 13

30
 Table II: Dynamics of Issuance
This table presents the issuance patterns for both bunching and non-bunching (control) issuers.
The variable Number of Issuances denotes the average number of times an issuer raised bank
qualified debt during the 2000-2008 period. Years between Issuances represents the average
time interval between consecutive debt issuances for issuers that raised debt more than once
within this time period. Standard deviations are provided in parentheses.

(1) (2) (3)

Bunching Issuers Controls Diff.

Number of Issuances 2.798 1.767 1.032***

(1.802) (1.256) std. er.. = 0.009

Years between Issuances 1.789 2.059 -0.270***

(1.232) (1.397) std. err.= 0.023

31
 Table III: Pre-treatment characteristics
Panel A in this table shows local governments’ budgets for a sample of hand-matched munici-
palities. Budget characteristics are as of the 2007 Census. Bunching and non-bunching qualified
issuers are defined in Section V.A. Panel B reports pre-trends for treated (Intensity > 0) and
control (Intensity = 0) counties. Column (5) reports differences in estimated coefficients af-
ter controlling for population decile by year fixed effects. Robust standard errors clustered
by county are reported in parenthesis. Significance follows: * 10 percent; ** 5 percent; *** 1
percent.

(1) (2) (3) (4) (4)

Bunching Non-Bunching Difference St. Err. Obs.
Population 24,817 7,975 16,842.226*** (1,187.758) 3,029
Revenues ($ ths) 43,902 12,962 30,939.648*** (2,626.351) 3,029
Intergv’t Rev. ($ths) 7,312 1,946 5,366.176*** (638.158) 3,029
Taxes ($ths) 16,043 4,538 11,505.068*** (869.303) 3,029
Property Taxes ($ths) 9,161 2,624 6,536.932*** (666.735) 3,029
Expenditures ($ths) 45,396 13,483 31,912.863*** (2,565.879) 3,029
Debt Outstanding ($ths) 43,856 13,438 30,418.511*** (3,208.893) 3,029

Revenues (per cap.) 1.965 1.804 0.161 (0.100) 3,029

Intergv’t Rev. (per cap.) 0.289 0.292 -0.002 (0.024) 3,029
Taxes (per cap.) 0.722 0.592 0.130** (0.060) 3,029
Property Taxes (per cap.) 0.428 0.372 0.055 (0.051) 3,029
Expenditures (per cap.) 2.053 1.961 0.092 (0.107) 3,029
Debt Outstanding (per cap.) 2.236 1.941 0.295 (0.184) 3,029

(1) (2) (3) (4) (5) (6) (7)

Treated Control Diff. St. Err. Diff. (adj.) St. Err. Obs.

∆ Employment 0.007 0.005 0.002 (0.002) 0.003 (0.002) 1,664

∆ Private Emp 0.006 0.005 0.001 (0.002) 0.002 (0.003) 1,660
∆ Gvt Emp 0.011 0.009 0.003 (0.002) 0.003 (0.003) 1,475
∆ Goods Emp -0.014 -0.006 -0.008* (0.005) -0.004 (0.005) 1,660
∆ Service Emp 0.012 0.009 0.003 (0.002) 0.005* (0.003) 1,664
∆ Wages 0.045 0.044 0.001 (0.003) 0.003 (0.004) 1,664
∆ Private Wages 0.045 0.046 -0.001 (0.003) 0.001 (0.004) 1,660
∆ Gvt Wages 0.046 0.045 0.001 (0.004) 0.001 (0.004) 1,516
∆ Goods Wages 0.032 0.040 -0.008 (0.006) -0.006 (0.007) 1,660
∆ Service Wages 0.051 0.049 0.002 (0.003) 0.005 (0.004) 1,664
∆ HPI 0.020 0.027 -0.007*** (0.002) -0.002 (0.003) 1,664
∆ Issuance 0.003 0.015 -0.012 (0.036) -0.013 (0.041) 1,664

32
 Table IV: Summary Statistics
This table presents summary statistics for the key variables used in the analysis. Intensity
refers to the fraction of bunching municipalities within a county affected by the policy change,
as detailed in Section V.A. Issuance represents the total amount of bank-qualified debt issued
by both bunching and control issuers in a year. Data on issuance, employment and wages span
the years 2004-2010.

(1) (2) (3) (4)

Obs. Mean St.Dev. Median

Intensity 2,912 0.136 0.121 0.115

Issuance ($ Mils.) 2,912 23.48 35.10 14.22
Employment - Total 2,912 137,298 305,732 46,486
Employment - Private 2,912 116,863 265,280 38,336
Employment - Goods 2,912 22,263 45,759 8,161
Employment - Services 2,912 94,653 221,001 28,686
Employment - Local Gvt 2,912 13,157 30,620 5,002
Wages - Total ($Bils.) 2,912 6.180 15.46 1.585
Wages - Private ($Bils.) 2,912 5.253 13.41 1.289
Wages - Goods ($Bils.) 2,912 1.257 2.902 0.370
Wages - Service ($Bils.) 2,912 3.998 10.70 0.839
Wages - Local Gvt ($Bils.) 2,912 0.561 1.552 0.165

33
 Table V: Real Effects of Bank Qualification - Issuance
This Table reports first stage results from the 2SLS specification in Eq. 6. Intensity is the
fraction of bunching municipalities in a county. P ost takes value of one during the regulatory
change (2009-2010), and zero otherwise (2004-2008). Issuance is the bank-qualified debt raised
in a county each year as specified in Section V.A, and expressed in $100Mils. Controls include
HPI and population. Robust standard errors clustered by county are reported in parenthesis.
Significance follows: * 10 percent; ** 5 percent; *** 1 percent.

(1) (2) (3)

Dep. Var.: Issuance Issuance Issuance

Intensity x Post 0.297*** 0.293*** 0.224***

(0.080) (0.080) (0.072)
Post 0.012 0.015
(0.010) (0.010)

Controls No Yes Yes

County FE Yes Yes Yes
Size Decile x Year FE No No Yes

Observations 2,912 2,912 2,912

Adjusted R-squared 0.828 0.829 0.835

34
 Table VI: Real Effects of Bank Qualification - Employment
This Table reports second stage results from the 2SLS specification in Eq. 7. The dependent variable is the log of employment.
Intensity is the fraction of bunching municipalities in a county. P ost takes value of one during the regulatory change (2009-2010),
and zero otherwise (2004-2008). Issuance is the bank qualified debt (expressed in $100Mils) raised in a county each year as
specified in Section V.A, and is instrumented using Intensityi x P ostt . Controls include HPI and population. Robust standard
errors clustered by county are reported in parenthesis. Significance follows: * 10 percent; ** 5 percent; *** 1 percent.

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Dep. Var.: Log(Employment) Log(Private Employment) Log(Gvt Employment)

Issuance 0.188** 0.158* 0.193* 0.231** 0.199** 0.219* 0.048 0.019 0.058

35
(0.087) (0.082) (0.116) (0.099) (0.094) (0.131) (0.091) (0.090) (0.107)
Post -0.041*** -0.040*** -0.055*** -0.053*** 0.015** 0.016**
(0.004) (0.004) (0.005) (0.005) (0.008) (0.008)

Controls No Yes Yes No Yes Yes No Yes Yes

County FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Size Decile x Year FE No No Yes No No Yes No No Yes

F-stat 1st stage 13.78 13.35 9.621 13.66 13.23 9.542 15.83 15.35 12.88

Observations 2,912 2,912 2,912 2,905 2,905 2,905 2,622 2,622 2,622
Counties 416 416 416 415 415 415 382 382 382
 Table VII: Private Sector Employment- Goods vs. Services
This Table reports second stage results from the 2SLS specification in Eq. 7. The dependent
variable is the log of private employment, broken down in goods and service sectors. Issuance
is the instrumented bank qualified debt (expressed in $100Mils) raised in a county each year as
specified in Section V.A. Controls include HPI and population. Robust standard errors clustered
by county are reported in parenthesis. Significance follows: * 10 percent; ** 5 percent; *** 1
percent.

(1) (2)
Dep. Var.: Log(Goods Emp) Log(Service Emp)

Issuance -0.232 0.390**

(0.189) (0.170)

Controls Yes Yes

County FE Yes Yes
Size Decile x Year FE Yes Yes

F-stat 1st stage 9.542 9.621

Observations 2,905 2,912

Counties 415 416

36
 Table VIII: Excluding High Foreclosure States
This Table reports second stage results from the 2SLS specification in Eq. 7. This sample
excludes the states most affected by the foreclosure crisis. Issuance is the instrumented bank
qualified debt (expressed in $100Mils) raised in a county each year as specified in Section V.A.
Controls include HPI and population. Robust standard errors clustered by county are reported
in parenthesis. Significance follows: * 10 percent; ** 5 percent; *** 1 percent.

(1) (2) (3)

Dep. Var.: Log(Emp) Log(Priv. Emp) Log(Gvt Emp)

Issuance 0.190* 0.221* 0.044

(0.115) (0.130) (0.105)

Controls Yes Yes Yes

County FE Yes Yes Yes
Size Decile x Year FE Yes Yes Yes

F-stat 1st stage 10.11 10.03 13.39

Observations 2,821 2,814 2,533

Counties 403 402 369

37
 Table IX: Bank Holdings of Municipal Bonds
This Table reports results from the specification in Eq. 9. Log(M uniBonds) is the log of mu-
nicipal bond holdings, measured at the bank holding company level. BankIntensity is the stan-
dardized bank’s exposure to municipal bond issuance as defined in Eq. 8. N etIncomeRatiot−1 ,
Costof Depositst−1 , EquityRatiot−1 , CashRatiot−1 are measured at the bank holding company
level and are respectively net income over total assets, interest expense on deposits over de-
posits, equity over total assets, and cash over total assets, each lagged by one year. Robust
standard errors are clustered at the bank level and are reported in parenthesis. Significance
follows: * 10 percent; ** 5 percent; *** 1 percent.

(1) (2) (3)

Dep. Var.: Log(Muni Bonds) Log(Muni Bonds) Log(Muni Bonds)

Bank Intensity 0.095* 0.094* 0.071**

(0.053) (0.050) (0.036)
N etIncomeRatiot−1 0.424* 0.188
(0.219) (0.197)
Costof Depositst−1 -0.061 -0.054
(0.047) (0.043)
EquityRatiot−1 -2.568***
(0.371)
CashRatiot−1 1.074***
(0.413)

Bank FE Yes Yes Yes

Observations 25,683 25,379 25,379

Adjusted R-squared 0.841 0.843 0.843

38
 Table X: Private Lending
This Table reports results from the reduced form specification 10. Log(CRALoans) is the
log of a bank’s new CRA loans extended in a county. Log(HM DAOrigin) is the log of a
bank’s new mortgages originated in a county. Log(HM DAP urchased) is the log of a bank’s
new mortgages purchased in a county. BankIntensity is the standardized bank’s exposure
to municipal bond issuance as defined in Eq. 8. N etIncomeRatiot−1 , Costof Depositst−1 ,
EquityRatiot−1 , CashRatiot−1 are respectively net income over total assets, interest expense
on deposits over deposits, equity over total assets, and cash over total assets, each lagged by
one year. Variables are measured at the bank holding company level. Robust standard errors
are double clustered by county and bank, and are reported in parenthesis. Significance follows:
* 10 percent; ** 5 percent; *** 1 percent.

(1) (2) (3)

Dep. Var.: Log(CRA Loans) Log(HMDA Origin) Log(HMDA Purchased)

Bank Intensity -0.187 -0.189* 0.102

(0.620) (0.108) (0.162)
N etIncomeRatiot−1 2.691 0.268 2.572
(2.098) (1.900) (2.499)
Costof Depositst−1 4.899*** 6.844 -7.639
(1.077) (5.985) (9.010)
EquityRatiot−1 0.032 2.086 -3.379
(2.931) (2.266) (3.287)
CashRatiot−1 -1.816 -3.424 4.155
(2.150) (2.724) (3.333)

Bank FE Yes Yes Yes

County x Year FE Yes Yes Yes

Observations 281,469 428,215 428,215

Adjusted R-squared 0.315 0.396 0.660

39
 .06
Not Qualified

.05
Bank Qualified

.04
Holdings/Assets
.02 .03
.01
0

2000 2002 2004 2006 2008 2010

Figure 1. Banks’ Holdings of Municipal Bonds.

This figure plots average banks’ holdings of bank-qualified (solid line) and non-qualified (dashed
line) municipal bonds, as of December of the calendar year. Holdings are expressed as fraction
of total assets. Data comes from Call Reports.

40
 3000
2000
Number of Issuers
1000
0

0 10 20 30 40 50
Issuance (mils)

(a) Panel A: Issuance Distribution, 2000-2008

800
600
Number of Issuers
400
200
0

0 10 20 30 40 50
Issuance (mils)

(b) Panel B: Issuance Distribution, 2009-2010

Figure 2. Panel A plots the distribution of issuers for years 2000-2008. Panel B plots the
distribution of issuers for years 2009-2010. On the x-axis is the size of the debt issuance per
year, broken down in $500k increments, while on the y-axis is the number of municipal issuers
in each size bin.

41
 Jump in Spread at Bank Qualification Limit
.2
.15
Difference in Spread
.1
.05
0
-.05
-.1

5 6 7 8 9 10 11 12 13 14 15
Issuance (mils)

Figure 3. Municipal Bond Spreads.

This figure shows the jump in spreads around the bank qualification cut-off. The sample includes
general obligation bonds issued before 2008. Spreads are defined as tax-adjusted municipal bond
yields over maturity-matched treasuries. On the x-axis is the size of the debt issuance. The
coefficients are expressed as difference from the average spread of $10M issuances. The dashed
lines represent 90% confidence intervals.

42
 8
Empirical 2009-2010
Counterfactual
7

5
Issuers (%)

0
0 10 20 30 40 50 60 70 80 90 100
Municipal Issuance ($M)

Figure 4. Counterfactual Validation.

This figure plots the counterfactual distribution obtained using the bunching estimation de-
scribed in Section IV (in red) along with the observed issuance distribution for the period
2009-2010 (in black). The x-axis represents issuance. The y-axis is the fraction of issuers in
each $500k issuance size bin.

43
 3000
Empirical
Counterfactual
2500

2000
Number of Issuers

1500

1000

500

0
0 5 10 15 20 25 30 35
Municipal Issuance ($M)

Figure 5. Bunching Estimation

This figure plots the observed distribution of issuers (solid black line) alongside the estimated
counterfactual (dashed red line) for the years 2000-2008, around the bank qualification cut-off.
The estimation follows the methodology described in Section IV. Each point represents the
count of issuers in each 5% issuance size bin, while the vertical dashed line marks the region of
exclusion.

44
 80
70 60
Percent of Issuers
30 40 20
10
0 50

0 1 2 3 4 5 6 7 8 9 10
BQ Issuance Size

(a) Panel A: Issuance Distribution at t + 1 - Bunching

80
70 60
Percent of Issuers
30 40 20
10
0 50

0 1 2 3 4 5 6 7 8 9 10
BQ Issuance Size

(b) Panel B: Issuance Distribution at t + 1 - Controls

Figure 6. This figure shows the distribution of issuance for bunching (Panel A) and non-
bunching (Panel B) issuers, conditional on having issued the year before.

45
 3000

1 2 3 4
2000
Number of Issuers
1000
0

0 10 20 30 40
Issuance (mils)

Figure 7. Issuers Classification.

This figure illustrates the classification of issuers across regions of the issuance distribution as
described in Section V.A

46
 BQ Issuance
.5
.4
Coefficient on Treated
.3
.2
.1
0
-.1
-.2
-.3

2004 2005 2006 2007 2008 2009 2010

Years

Figure 8. Effects on Issuance.

This figure illustrates the effects on municipal bond issuance using a calendar-time coefficient
plot. The plot shows the regression coefficients of the logarithm of Issuance on the interaction
between treated Intensity and year dummies, covering the period from 2004 to 2010. The year
2008 serves as the base year. The dashed lines correspond to 90% confidence intervals.

47
 Local Gvt Employment

.08
.03
Coefficient on Treated
-.02
-.07
-.12

2004 2005 2006 2007 2008 2009 2010

Years

(a) Panel A: Government Employment

Private Employment
.11
.06
Coefficient on Treated
.01
-.04
-.09
-.14

2004 2005 2006 2007 2008 2009 2010

Years

(b) Panel B: Private Employment

Private Emp.- Service
.11
.06
Coefficient on Treated
.01
-.04
-.09
-.14

2004 2005 2006 2007 2008 2009 2010

Years

(c) Panel C: Private Employment - Services

Figure 9. This figure illustrates the effects on employment using a calendar-time coefficient
plot. The plot shows the regression coefficients of the logarithm of employment by sector on
the interaction between treated Intensity and year dummies, covering the period from 2004 to
2010. The year 2008 serves as the base year. The dashed lines correspond to 90% confidence
intervals.

48
 Change in House Price Index
.04
.03
.02
Coefficient on Treated
.01
0
-.04 -.03 -.02 -.01

2005 2006 2007 2008 2009 2010

Years

Figure 10. House Price Index.

This figure reports the regression coefficients of the logarithm of the house price index on the
interaction between treated and year dummies, covering the period from 2004 to 2010, with
2004 as the base year.

49
 Change in Municipal Bonds Holdings
-.5 -.3 -.1 .1 .3 .5 .7 .9

-.5 -.3 -.1 .1 .3 .5 .7 .9

-.5 -.3 -.1 .1 .3 .5 .7 .9

2005 2006 2007 2008 2009 2010 2005 2006 2007 2008 2009 2010 2005 2006 2007 2008 2009 2010
Bank Size Quintile = 1 Bank Size Quintile = 2 Bank Size Quintile = 3
-.5 -.3 -.1 .1 .3 .5 .7 .9

-.5 -.3 -.1 .1 .3 .5 .7 .9

2005 2006 2007 2008 2009 2010 2005 2006 2007 2008 2009 2010
Bank Size Quintile = 4 Bank Size Quintile = 5

Figure 11. Banks’ Change in Muni Bond Holdings by Quintile.

This figure plots the regression coefficients of the logarithm of municipal bond holdings against
year dummies for the period 2005-2010, with 2008 as the base year, broken down by banks’
asset size quintiles. The shaded areas correspond to 90% confidence intervals.

50
 Appendix

I. Qualified Small Issuer

A municipality receives the qualified small issuer designation if it can reasonably expect to
issue within the calendar year tax-exempt obligations within the limit of $10M. This limit was
extended to $30M in 2009-2010. Private activity bonds, that is bonds that do not pass the public
purpose test, are excluded from being designated as qualified regardless of size. Exceptions are
provided for 501(c)(3) types under the Internal Revenue Code Section 145.

Conduit bonds under 501(c)(3) count towards the qualification limit of the borrower that
issued them. In other words, a municipality and the entities or authorities that issue on its
behalf count as one issuer, and their aggregate issuance needs to remain within the $10M limit
(aggregation).

Refunding obligations that do not exceed the obligation they purport to refund are generally
not qualifiable, unless acting in the form of advanced refunding. However, a refunding obligation
issued to refund a designated qualified obligation outstanding is allowed to be designated as
qualified itself, to the extent that the average maturity date of the refunding obligation does
not exceed the average maturity of the bonds it stands to refund.

II. Banks’ Taxation of Municipal Bonds

Banks deduct interest expense in their income statement, and pay taxes on the profits which
are calculated as of net of the deductions. Higher deductions imply a lower taxable base. The
Tax Reform Act of 1986 however provides that “no deduction shall be allowed for interest
on indebtedness incurred or continued to purchase or carry obligations the interest on which is
wholly exempt from tax” ( §265(a)).This means that 100% of a bank’s interest expense incurred
to enter a position into tax-exempt income, is not allowed to be deducted. The bank is hence
penalized for acquiring a position in a tax-exempt asset through this deduction disallowance,
also known as pro-rata disallowance.

The Act, §265(b)(3), however also provides an exception for qualified tax-exempt obligations,
also known as bank qualified bonds. Bank Qualified municipal bonds are subject to a lenient
treatment: only 20% of the interest cost incurred cannot be deducted (compared to 100% for
non-qualified bonds). In other words, banks can shield from taxation 80% of the carrying cost

51
of a Bank Qualified obligation.

The end-of-year pro-rata disallowance is so calculated:

Tax Exempt Obligations
× Year-to-Date Interest Expense × D
All Assets

where D is the percentage disallowed: 20% for Bank-Qualified bonds, and 100% for non-
Bank qualified bonds.

It is worth noting that the Tax Reform Act of 1986 also affected insurers. However, the tax
treatment of insurance companies does not embed a discontinuity. Specifically, the insurance
sectors is subject to a proration provision that adds 15% of tax-exempt income back into their
regular taxable income. In other words, the effective after tax yield, adjusted for proration, that
an insurance company earns on a tax-exempt municipal bond is equivalent to the unadjusted
yield multiplied by a factor of: (1 − 15%τ ), where τ is the insurer tax rate. This is important
to note since it implies that the $10M cutoff uniquely identifies segmentation at the level of
bank financing.

III. Backing out Banks’ Qualified Holdings from Call Reports

Banks’ balance sheets report the aggregate holdings of municipal securities available for sale
or held to maturity. Holdings of municipal securities include both qualified obligations and
non-qualified obligations. Loans and leases to States and Local governments are also reported
on Call Reports.

The memorandum item to the income statement 4513, however, requires banks to file interest
expenses incurred to carry tax-exempt municipal securities and loans, with the exclusion of
bank-qualified tax exempt obligation. The item requires banks to report the following dollar
value, as of end of December for the entire calendar year:
Non-Qualified Tax Exempt Securities + Loans &Leases
×Year-to-Date Total Interest Expense
Total Assets

Total Assets and Total Interest Expenses are reported in the balance sheets, as well as loans
and leases to municipalities. It is therefore possible to calculate the bank’s exposure to non
qualified municipal securities from the memorandum. The qualified holdings are then obtained
by subtracting the non-qualified holdings from the aggregate municipal securities holdings.

52
IV. A Simple Model of Local Debt Financing

Consider a myopic politician in municipality i who derives utility from maximizing govern-
ment expenditures, while bearing a cost of debt issuance:

!1+ α1
ζi B
Ui (G, B) = G − 1 (A1)
1+ α
ζi

Local government expenditure is denoted by G, while the cost of issuing a bond of size
B is represented by the second term, where ζi captures the financing needs of municipality i,
and α is the elasticity of the cost of issuance. The larger the bond, the higher the cost of
issuance. This cost can be broadly intended as book-building expenses, which increase with the
size and complexity of the issue, or political costs of approving a large debt issuance, as well as
reputational cost of incurring extensive debt.18

The politician faces the following budget constraint:

G = B + Π − rB (A2)

Local government expenses are sustained by bond issuance – net of end-of-period interest
repayment, rB, – and taxes, Π. This set-up can be interpreted as a reduced form for a multi-
period budget constraint where rB would show up next period. The amount of taxes levied in
the period is assumed to be exogenous. While municipalities are not technically constrained by
balanced budget provisions, there are statutory limits on tax hikes.

From the F.O.C. for maximization of equation (A1) subject to (A2) we obtain the optimal
bond supply:
B S = ζi (1 − r)α (A3)

that is, bond issuance depends on the debt needs of municipality i and on the equilibrium
interest rate, r, with elasticity, α. Heterogeneity across municipalities is driven by the financing
needs, ζi , which are assumed to be distributed smoothly and with density f (ζ).

Demand is aggregated across two types of investors: households and banks. Both households
and banks have a simple demand function for local government bonds, specifically demand is
linear in the after-tax interest rate. For households the return is tax-free, so the after-tax and
18
Majority voting is occasionally required before issuing a general obligation bond.

53
the pre-tax interest rates coincide:
BH = β r (A4)

On the contrary, bank demand is proportional to the taxation schedule they face: any bond of
size smaller than B ∗ is subject to tax rate t, whereas a bond of size B > B ∗ is taxed at rate
t + ∆t:
B B = [1 − (t + ∆t 1{B > B ∗ })] r (A5)

This discontinuity in taxation for banks captures the bank qualification regulation. Equating
bond demand and supply, the interest rate on the municipal bond issued solves:
(1 − r)α 1−t+β
= (A6)
r ζi
As ζ,α, and β are fixed parameters, this implies that when there is a jump in the taxation
schedule of banks, which increases from t to t + ∆t on the entirety of the bond issue, then the
interest rate on the bond increases. In other words, defining t = t0 , and t1 = (t0 + ∆t0 ) 1{B >
B ∗ }, it follows that:
r(t1 ) > r(t0 ) (A7)

The budget constraint that the politician in municipality i faces (equation (A2)), can then
be rewritten as 
Π + B(1 − r(t0 )), if B ≤ B ∗


G= (A8)

Π + B(1 − r(t1 )), otherwise


The budget constraint thus exhibits a jump at B ∗ , as represented in Figure A2 Panel A.

When faced with the notch, the municipality that would have otherwise issued B ∗ + ∆B ∗ , is
indifferent between locating at B I and B ∗ , and chooses to bunch at the threshold. Consider
the case of quadratic issuance costs, then the distribution of debt issuance in the presence of a
notch, H1 (B), is such that:

(1−t +β)B ∗
ζi (1 − r(t0 )) if ζi < 1−t00+β−B ∗





 h i
(1−t0 +β)B ∗ (1−t1 +β)B ∗
B = B∗ if ζi ∈ 1−t ∗ , ∗
(A9)
 0 +β−B 1−t1 +β−B



ζi (1 − r(t1 )) if ζi > (1−t1 +β)B ∗∗


1−t1 +β−B

In words, under a smooth distribution of financing needs, ζi , aggregating across municipali-

54
ties generates an excess mass at B ∗ , as well as a missing mass of municipalities to the immediate
right of the qualification limit. Panel B in Figure A2 shows the effect of the notch on the density
of issuance, with the dotted line representing the counterfactual distribution, h0 (B). The mass
of municipalities bunching at the limit is therefore given by
Z B ∗ +∆B ∗
D= h0 (B)dh ≈ h0 (B ∗ )∆B ∗ (A10)
B∗

where the approximation follows from the assumption of a constant counterfactual distribution
in the interval [B ∗ , B ∗ + ∆B ∗ ]. ∆B ∗ represents the quantity of interest, that is the behavioral
response of municipalities generated by the bank-qualification rule.

The model can be extended along many dimensions, such as allowing for heterogeneity.
The theoretical take-away remains valid also under the extensions: in such case, the mass of
bunching municipalities would estimate the average response across marginal bunching cities
associated with each elasticity and demand parameters.19 In the framework of the extended
model, this would be:
Z Z Z B ∗ +∆B(β,α)
∗

h0 (B)dh dβ dα ≈ h0 (B ∗ )E[∆Bα,β
∗
]
α β B∗

This simple model formalizes the evidence presented in Section IV.A, and demonstrates how
a discontinuity in the tax schedule for a subset of municipal investors generates a notch in the
budget constraints of the local government, which in turn affects the municipality’s decision to
raise capital and leads it to reduce its issuance and bunch at the bank qualification cut-off.

19
See Kleven and Waseem (2013) for in-depth theoretical discussion in the context of income tax notches.

55
 Table A1: Sample Restrictions
This table outlines the impact of each sampling restriction on the sample used in the paper,
detailing how the final sample compares to the universe of counties in terms of total population
and employment. It also shows the proportion of the U.S. population and employment repre-
sented by the final sample.

Population Employment
(1) (2) (3) (4) (5) (6) (7)
Sample: # Counties Mean St.Dev. Frac. Mean St.Dev. Frac.

Universe 3,084 97,404.4 309,400.0 1 42,217.0 148,342.7 1

Issued BQ Bonds 2,199 113,078.4 348,990.8 0.83 49,217.7 162,872.5 0.83
Present Every Year 477 262,906.1 621,097.2 0.42 120,433.5 288,947.5 0.44
Only Urban 416 299,363.1 657,297.9 0.41 137,298.2 305,830.3 0.44

56
 Table A2: Distribution of Counties
This table reports the distribution of treated (bunching) and control counties in the sample
across U.S. states.

State Counties (#) Counties (%) Bunching Control

Alabama 2 0.48 1 1
Arizona 2 0.48 2 0
Arkansas 9 2.16 8 1
California 9 2.16 8 1
Colorado 2 0.48 2 0
Connecticut 7 1.68 6 1
Illinois 31 7.45 20 11
Indiana 3 0.72 3 0
Iowa 17 4.09 13 4
Kansas 10 2.4 9 1
Kentucky 4 0.96 0 4
Louisiana 1 0.24 1 0
Maine 1 0.24 1 0
Massachusetts 10 2.4 9 1
Michigan 10 2.4 9 1
Minnesota 34 8.17 27 7
Missouri 21 5.05 17 4
Montana 1 0.24 0 1
Nebraska 11 2.64 9 2
New Hampshire 1 0.24 1 0
New Jersey 20 4.81 18 2
New Mexico 4 0.96 3 1
New York 42 10.1 33 9
North Dakota 1 0.24 1 0
Ohio 33 7.93 26 7
Oklahoma 28 6.73 7 21
Oregon 1 0.24 1 0
Pennsylvania 31 7.45 30 1
South Carolina 1 0.24 1 0
South Dakota 1 0.24 1 0
Tennessee 1 0.24 1 0
Texas 22 5.29 21 1
Utah 2 0.48 2 0
Washington 6 1.44 5 1
Wisconsin 37 8.89 23 14

57
 Table A3: Alternative Samples
This Table reports second stage results from the 2SLS specification in Eq. 7, under alternative definitions of treated (bunching)
and controls. The dependent variable is the log of total employment, broken down by sector. Intensity is the fraction of bunching
municipalities in a county. P ost takes value of one during the regulatory change (2009-2010), and zero otherwise (2004-2008).
Issuance is the bank qualified debt (expressed in $100Mils) raised in a county each year as specified in Section V.A, and is
instrumented using Intensityi x P ostt . Controls include HPI and population. Robust standard errors clustered by county are
reported in parenthesis. Significance follows: * 10 percent; ** 5 percent; *** 1 percent.

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Dep. Var.: Log(Employment) Log(Private Employment) Log(Gvt Employment)

Issuance 0.217* 0.224* 0.213* 0.256* 0.265* 0.245* 0.038 0.025 0.050

58
(0.118) (0.127) (0.117) (0.134) (0.144) (0.135) (0.102) (0.104) (0.099)

Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes
County FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Size Decile x Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes

F-stat 1st stage 10.37 9.652 7.820 10.30 9.599 7.745 13.85 13.43 11.82

Bunching Region ($ M) (8.5,10] (8.5,10] [9,10] (8.5,10] (8.5,10] [9,10] (8.5,10] (8.5,10] [9,10]
Control Region ($ M) (0,8.5] (0,8) (0,9) (0,8.5] (0,8) (0,9) (0,8.5] (0,8) (0,9)

Observations 3,339 3,255 3,304 3,328 3,244 3,293 3,008 2,938 2,973
Counties 477 465 472 476 464 471 440 430 435
 Table A4: Use of Funds
This table reports the breakdown of the use of proceeds from bank-qualified bonds issued in
2009-2010.

Use of Funds Bonds (#) Bonds (%)

Airports 45 0.11
Bridges 21 0.05
Civic & Convention Centers 2 0
Combined Utilities 409 1
Communication 9 0.02
Correction Facs/Courts 104 0.25
Econ Dev 145 0.35
Electric & Public Power 188 0.46
Fire Station & Equipment 377 0.92
Flood Control/Storm Drain 158 0.39
Gas 25 0.06
Gen Purpose/Pub Impt 16982 41.48
Govt/Pub Buildings 293 0.72
Higher Education 557 1.36
Hospital Equipment Loans 5 0.01
Hospitals 76 0.19
Libraries & Museums 189 0.46
Lifecare/Retire Centers 42 0.1
Mass/Rapid Transit 26 0.06
Multi-Family Housing 63 0.15
Nursing Homes 3 0.01
Other Education 7 0.02
Other Recreation 145 0.35
Other Transportation 32 0.08
Parking Facilities 33 0.08
Parks, Zoos & Beaches 585 1.43
Police Station & Equipment 41 0.1
Primary & Secondary Education 13572 33.15
Public Safety 59 0.14
Sanitation 548 1.34
Solid Waste/Resource Recovery 7 0.02
Stadiums/Sports Complex 34 0.08
Toll Roads/Highways/Streets 691 1.69
Water & Sewer 5472 13.36

59
 Table A5: Real Effects of Bank Qualification - Wages
This Table reports second stage results from the 2SLS specification in Eq. 7. The dependent variable is the log of total wages,
broken down by sector. Intensity is the fraction of bunching municipalities in a county. P ost takes value of one during the
regulatory change (2009-2010), and zero otherwise (2004-2008). Issuance is the bank qualified debt (expressed in $100Mils)
raised in a county each year as specified in Section V.A, and is instrumented using Intensityi x P ostt . Controls include HPI and
population. Robust standard errors clustered by county are reported in parenthesis. Significance follows: * 10 percent; ** 5
percent; *** 1 percent.

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Dep. Var.: Log(Wages) Log(Private Wages) Log(Gvt Wages)

Issuance 0.151 0.073 0.153 0.187 0.105 0.190 0.115 0.056 0.105

60
(0.106) (0.095) (0.133) (0.120) (0.107) (0.151) (0.128) (0.128) (0.142)
Post 0.051*** 0.059*** 0.033*** 0.041*** 0.126*** 0.129*** -
(0.006) (0.006) (0.007) (0.006) (0.011) (0.011)

Controls No Yes Yes No Yes Yes No Yes Yes

County FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Size Decile x Year FE No No Yes No No Yes No No Yes

F-stat 1st stage 13.78 13.35 9.621 13.66 13.23 9.542 15.91 15.46 12.94

Observations 2,912 2,912 2,912 2,905 2,905 2,905 2,663 2,663 2,663
Counties 416 416 416 415 415 415 384 384 384
 Table A6: Issuance - Large Issuers
This Table reports first stage results from the 2SLS specification in Eq. 6. Intensity is the
fraction of Zone (3) municipalities in a county. P ost takes value of one during the regulatory
change (2009-2010), and zero otherwise (2004-2008). Issuance is the municipal debt raised in
a county each year, and expressed in $100Mils. Controls include HPI and population. Robust
standard errors clustered by county are reported in parenthesis. Significance follows: * 10
percent; ** 5 percent; *** 1 percent.

(1) (2) (3)

Dep. Var.: Issuance Issuance Issuance

Intensity x Post -0.871 -0.338 0.154

(0.694) (0.660) (0.610)
Post -0.156 -0.132
(0.107) (0.109)

Controls No Yes Yes

County FE Yes Yes Yes
Size Decile x Year FE No No Yes

Observations 3,766 3,766 3,766

Adjusted R-squared 0.825 0.827 0.832

61
 Table A7: Bank Characteristics
This table reports summary statistics of bank characteristics for banks above and below median
values of BankIntensity. Balance sheet values are measured at the bank holding company level
and are expressed in millions.

Panel A: Above Median BankIntensity

(1) (2) (3) (4)
Obs. Mean St.Dev. Median

Assets 13,062 179.3 307.8 97.53

Loans 13,062 122.6 214.8 64.06
Deposits 13,062 147.5 247.7 81.52
TRSY & Agency debt 13,062 12.91 31.37 5.513
Muni Bonds 13,062 7.026 15.40 1.854
Equity Ratio 13,062 0.113 0.0624 0.0995
Net Income Ratio 13,062 0.000333 0.0359 0.00604
Cost of Deposits 13,061 0.0831 0.0355 0.0798
Cash Ratio 13,062 0.0660 0.0734 0.0422

Panel B: Below Median BankIntensity

(1) (2) (3) (4)
Obs. Mean St.Dev. Median

Assets 13,064 4,255 58,329 252.9

Loans 13,064 2,381 29,867 166.1
Deposits 13,064 2,249 25,873 208.8
TRSY & Agency debt 13,064 102.5 1,145 13.81
Muni Bonds 13,064 26.04 70.89 8.463
Equity Ratio 13,064 0.103 0.0387 0.0962
Net Income Ratio 13,064 0.00419 0.0251 0.00878
Cost of Deposits 13,063 0.0862 0.138 0.0811
Cash Ratio 13,064 0.0525 0.0516 0.0360

62
 Table A8: Treasury and Agency Debt
Log(T RSY ) is the log of a bank’s Treasury holdings. Log(Agency) is the log of a bank’s agency
debt holdings. BankIntensity is the standardized bank’s exposure to municipal bond issuance
as defined in Eq. 8. N etIncomeRatiot−1 , Costof Depositst−1 , EquityRatiot−1 , CashRatiot−1
are respectively net income over total assets, interest expense on deposits over deposits, equity
over total assets, and cash over total assets, each lagged by one year. Robust standard errors
are double clustered by county and bank, and are reported in parenthesis. Significance follows:
* 10 percent; ** 5 percent; *** 1 percent.

(1) (2)
Dep. Var.: Log(TRSY) Log(Agency)

Bank Intensity 0.039 -0.150

(0.026) (0.094)
N etIncomeRatiot−1 -0.083 0.418
(0.197) (0.262)
Costof Depositst−1 -0.166 -0.103
(0.330) (0.284)
EquityRatiot−1 -1.002*** 0.152
(0.375) (0.583)
CashRatiot−1 0.515 -2.977***
(0.526) (0.549)

Bank FE Yes Yes

Observations 25,379 25,379

Adjusted R-squared 0.627 0.652

63
 Table A9: Rotemberg Weights
This table reports statistics on the Rotemberg weights following Goldsmith-Pinkham et al.
(2020). Panel A reports statistics on negative and positive Rotemberg weights. Panel B reports
the correlations between the weights (αˆk ), the average bank qualified issuance (gk ), the just-
identified coefficient estimates (βˆk ), and the variation in deposit shares across banks (V ar(zk )).
Panel C reports the top five counties according to their Rotemberg weights. Municipal bond
issuance is expressed in thousands. Coefficients on gk in Panel C are divided by 106 for read-
ability.

Panel A: Negative and Positive Weights

Sum Mean Share

Negative -0.394 -0.0003 0.22
Positive 1.394 0.0011 0.78

Panel B: Correlations
αˆk gk βˆk V ar(zk )
αˆk 1
gk 0.2933 1
βˆk 0.0026 -0.013 1
V ar(zk ) 0.0164 0.3192 -0.0026 1

Panel C: Top 5 Rotemberg weight counties

α̂k gk β̂k
Dallas County - Texas 0.098 0.137 -0.735
Harris County - Texas 0.075 1.260 -2.721
Westchester County - New York 0.068 0.505 -0.284
Williamson County - Texas 0.053 0.192 -0.463
Los Angeles County - California 0.052 0.219 -3.565

64
 (33.3675,1497.221]
(12.9365,33.3675]
(4.541,12.9365]
[.155,4.541]
No data

Figure A1. Bank Qualified Issuance.

This figure shows the distribution of aggregate bank-qualified issuance per county for the years
2009-2010. Counties in progressively darker shades of blue correspond to the 0 to 25th, 25th
to 50th, 50th to 75th, and 75th to 100th percentile of the issuance distribution.

65
 Slope 1- r (t0)
G

Slope 1- r (t1)
C0

C1
C2

B* BI B*+ ΔB* B
Density

B* BI B*+ ΔB* B

Figure A2. Notch Theory.

This figure shows the impact of a tax-induced notch on a municipality’s budget set. The
notch represents a discrete jump in the average tax rate from t0 to t1 = t0 + ∆t in the bank’s
taxation schedule. When faced with the notch, a municipality that would have otherwise issued
B ∗ + ∆B ∗ , is indifferent between locating at B I and B ∗ , and chooses to bunch at the threshold
(top panel). All issuers initially located on (B ∗ , B ∗ + ∆B ∗ ) bunch at the notch. The figure in
the bottom panel shows the corresponding post-notch density distribution, under homogeneous
elasticity, which exhibits sharp bunching at B ∗ and zero mass in (B ∗ , B I ).

66