ESSAYS IN MACROECONOMICS

By

Timothy Moreland

A DISSERTATION

Michigan State University

in partial fulﬁllment of the requirements

Submitted to

for the degree of

Economics – Doctor of Philosophy

2021

ABSTRACT

ESSAYS IN MACROECONOMICS

By

Timothy Moreland

Chapter 1: Financial Consolidation and the Cyclicality of Corporate Financing

We study the impact of the concentration and complexity of the banking sector on ﬁrms’ ﬁnanc-
ing and investment behavior over the business cycle. We ﬁnd that, after the late 1990s, while debt
issuance remained procyclical for U.S. ﬁrms of all sizes, equity issuance and liquidity accumula-
tion switched from countercyclical to procyclical for small and medium-sized publicly-traded ﬁrms.
Using matched ﬁrm-bank data, we provide evidence that bank consolidation contributed to this
change. We rationalize these ﬁndings in a general equilibrium business cycle model. After bank
consolidation, the weakening in ﬁrms’ bargaining power and relational ties with banks enhances
ﬁrms’ precautionary demand for liquidity and equity issuance incentives following positive shocks.
The change in ﬁnancing behavior increases investment and employment sensitivity to aggregate
productivity shocks.

Chapter 2: Monetary Policy and Firm Heterogeneity: The Role of Leverage Since the

Financial Crisis

We study how leverage determines ﬁrm-level responses to monetary policy. Using both high-
frequency ﬁnancial market and quarterly investment data, we ﬁnd that the role of leverage in
monetary transmission changed around the ﬁnancial crisis of 2007-09. Firms with high leverage
were less responsive to monetary policy shocks in the pre-crisis period but have become more
responsive since the crisis. The higher responsiveness is driven by ﬁrms whose leverage is more
dependent on long-term debt, suggesting an outsize role for monetary policy aﬀecting long-term
funding conditions since the crisis. We also ﬁnd suggestive evidence for transmission through
changes in monetary policy uncertainty.

Chapter 3: The International Spillover Eﬀects of US Monetary Policy Uncertainty
An extensive literature studies the international transmission of US monetary policy surprises
(shifts in expected path of the policy rate).
In this paper we show that changes in uncertainty
around the expected path constitute an important additional dimension of spillover eﬀects to global
bond yields. In advanced countries, it is the term premium component of yields that responds to
uncertainty. We ﬁnd that this can be explained by an international portfolio balance mechanism.
In contrast, for emerging countries it is the expected component of yields that reacts to uncertainty.
This can be rationalized from a ﬂight to safety channel. We ﬁnd heterogeneity in the country-level
response to uncertainty only in emerging countries and it is driven by the degree of ﬁnancial
openness. Finally, equity markets in both advanced and emerging countries also respond to US
monetary policy uncertainty.

Copyright by
TIMOTHY MORELAND
2021

TABLE OF CONTENTS

LIST OF TABLES .
LIST OF FIGURES .

.
.

.
.

.
.

.
.

.
.

.
.

.
.

.
.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xi

CHAPTER 1 FINANCIAL CONSOLIDATION AND THE CYCLICALITY OF COR-

.

.

.

.

.

.

.
.

. .

. .

1.4 Robustness . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Introduction .

1.1
.
1.2 Empirical Evidence .

.
.
Firm-Level Data .
.

1.2.1
1.2.2 Lender Data . .

1.5 The Role of Financial Consolidation . . . . . . . . . . . . . . . . . . . . . . . .

PORATE FINANCING . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1
1
6
6
9
1.3 Firm Financing over the Business Cycle . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.1 Aggregate Time Series Evidence . . . . . . . . . . . . . . . . . . . . . . . 11
Firm-Level Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.2
. 15
1.4.1
Identifying the Structural Break . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.2 Other Robustness Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
. 18
1.5.1 Timing of Riegle-Neal Adoption . . . . . . . . . . . . . . . . . . . . . . . 18
1.5.2 Measures of Financial Consolidation . . . . . . . . . . . . . . . . . . . . . 19
. 22
1.6.1
Firm Technology and Financing . . . . . . . . . . . . . . . . . . . . . . . 22
1.6.2
Firm Decisions .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.6.3 Household Decisions and General Equilibrium . . . . . . . . . . . . . . . 28
. 29
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Simulated Responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
. 34
. 36

. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.8 Financial and TFP Shocks .
Investment and Employment
1.9
1.10 Conclusion .
.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.7 Calibration and Quantitative Analysis

. . . . . . . . . . . . . . . . . . . . . . .

1.7.1 Calibration .
1.7.2

1.6 The Model .

. .

.

.

.

.

.

.

.

.

.

.

.

.

.

CHAPTER 2 MONETARY POLICY AND FIRM HETEROGENEITY: THE ROLE OF

2.1
2.2 Data .

.

.
.

.
.

.
.

. .

. .

. .

.
. .

Introduction .

2.2.1 Monetary Policy Shocks
2.2.2

LEVERAGE SINCE THE FINANCIAL CRISIS . . . . . . . . . . . . . . . 38
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
. . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Firm-Level Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3.1 Evidence from ﬁrm-level stock returns . . . . . . . . . . . . . . . . . . . . 47
2.3.2 Leverage in the pre- and post-crisis samples . . . . . . . . . . . . . . . . . 50
2.3.3 Evidence from ﬁrm-level options data . . . . . . . . . . . . . . . . . . . . 52
2.3.4 Evidence from investment data . . . . . . . . . . . . . . . . . . . . . . . . 54

. .

. .

. .

.

.

.

2.3 Results .

v

.

.

.

. .

2.4 Mechanism . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.4.1 Ottonello & Winberry channels of monetary transmission . . . . . . . . .
. 57
2.4.2 Transmission through changes in long-term funding conditions . . . . . . . 62
2.4.3 Transmission through monetary policy uncertainty changes . . . . . . . .
. 64
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

2.5 Robustness Checks
2.6 Conclusion . .
. .

.
.

.
.

.
.

CHAPTER 3 THE INTERNATIONAL SPILLOVER EFFECTS OF US MONETARY

3.3 Data .

3.4 Results .

Introduction .

3.1
.
3.2 US Monetary Policy Shocks

. .

.

.

.

.

.

.

.

.

.

.

. .

. .

. .

. .

. .

. .

3.3.1

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

POLICY UNCERTAINTY . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
. 75
3.2.1 Monetary Policy Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.2.2 Monetary Policy Surprise . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.4.1 Response of US asset prices to US monetary policy uncertainty . . . . . . 80
3.4.2 Response of international bond yields to US monetary policy uncertainty . 81
3.4.3 Understanding the response . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.4.3.1 Response in advanced countries . . . . . . . . . . . . . . . . . . 89
3.4.3.2 Response in emerging countries . . . . . . . . . . . . . . . . . . 92
3.4.4 Heterogeneity in response to monetary policy uncertainty . . . . . . . . . . 94
3.4.5 Response of international equity indices to US monetary policy uncertainty
97
3.4.6 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
. .
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
.
SUPPLEMENTAL INFO FOR CHAPTER 1 . . . . . . . . . . . . . 103
TABLES FOR CHAPTER 1 . . . . . . . . . . . . . . . . . . . . . 106
FIGURES FOR CHAPTER 1 . . . . . . . . . . . . . . . . . . . . . 122
TABLES FOR CHAPTER 2 . . . . . . . . . . . . . . . . . . . . . 131
FIGURES FOR CHAPTER 2 . . . . . . . . . . . . . . . . . . . . . 145
SUPPLEMENTAL INFO FOR CHAPTER 3 . . . . . . . . . . . . . 147
TABLES FOR CHAPTER 3 . . . . . . . . . . . . . . . . . . . . . 149
FIGURES FOR CHAPTER 3 . . . . . . . . . . . . . . . . . . . . . 178
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
.

.
.

.
.

.
.

.
.

.

.

.

.

.

.

.

3.5 Conclusion .
.

.

.

APPENDICES .
APPENDIX A
APPENDIX B
APPENDIX C
APPENDIX D
APPENDIX E
APPENDIX F
APPENDIX G
APPENDIX H
BIBLIOGRAPHY .

vi

LIST OF TABLES

Table B1: Summary statistics .

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

Table B2: Cyclicality of Aggregate Financing Variables, by Size . . . . . . . . . . . . . . . 107

Table B3: Firm-Level Cyclicality of Financing Variables . . . . . . . . . . . . . . . . . . . 108

Table B4: State-Level Timing of Riegle-Neal Adoption . . . . . . . . . . . . . . . . . . . 109

Table B5: Reduction in Small Firm’s Bargaining Power, Syndicate Structure . . . . . . . . 110

Table B6: Reduction in Small Firm’s Bargaining Power, Lender Market Power

. . . . . . . 111

Table B7:

Increase in Small Firms’ Lender Relationship Complexity, MBHC Status,
1999-2012 .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

. .

. .

Table B8: Calibration .

.

.

.

.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

Table B9: Response of Financing Behavior to Positive TFP and Financial Shocks . . . . . . 114

Table B10: Firm-Level Cyclicality of Real Variables . . . . . . . . . . . . . . . . . . . . . . 115

Table B11: Robustness Tests: Firm Fixed Eﬀect and Consistent Sample

. . . . . . . . . . . 116

Table B12: Panel Regression: Alternative Financing Variables

. . . . . . . . . . . . . . . . 117

Table B13: Aggregate Financing Variables: No Filtering . . . . . . . . . . . . . . . . . . . 118

Table B14: Eﬀects of Consolidation by Size of Lender Pool, 1999-2012, All Lender Pool

. . 119

Table B15: Panel Regression: Financing Response to Negative Shock . . . . . . . . . . . . 120

Table B16: Panel Regression: Real Response to Negative Shock . . . . . . . . . . . . . . . 121

Table D1: Summary Statistics .

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

Table D2: Response of ﬁrm-level stock returns to monetary shocks

. . . . . . . . . . . . . 132

Table D3: Regression of ﬁrm-level implied volatility leading up to FOMC announcement

. 133

Table D4: Contemporaneous response of ﬁrm-level investment to monetary shocks . . . . . 134

vii

Table D5: Response of Bond Yield Spread to Monetary Policy Shock . . . . . . . . . . .

. 134

Table D6: Response of 10 year nominal yield, real yield and term premium to monetary

shocks

.

.

.

.

. .

. .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

Table D7: Response of ﬁrm-level stock returns (by high/low long-term leverage)

. . . . . . 136

Table D8: Response of ﬁrm-level stock returns to monetary policy uncertainty shocks

. . . 136

Table D9: Response of Firm-Level Implied Volatility to MPU Shock . . . . . . . . . . . . 137

Table D10: Summary Statistics of Firm Characteristics

. . . . . . . . . . . . . . . . . . . . 137

Table D11: Response of ﬁrm-level stock returns to monetary shocks w/ control interactions . 138

Table D12: Response of ﬁrm-level stock returns to monetary shocks: Without a time ﬁxed

eﬀect

.

.

.

.

.

. .

. .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

Table D13: Response of ﬁrm-level stock returns to monetary shocks (Pre. vs. Post 1SD

leverage outliers removed)

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

Table D14: Robustness of baseline results to alternative measure of leverage: Debt-to-Assets 141

Table D15: Robustness of baseline results to alternative measure of leverage: 1-quarter

lagged debt-to-capital . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

Table D16: Robustness of baseline results to time-sector FE . . . . . . . . . . . . . . . . .

. 143

Table D17: Robustness of baseline results to full CRSP/Compustat sample: No ﬁrm entry/exit144

Table G1: Summary Statistics .

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

Table G2: Response of US asset prices to monetary shocks

. . . . . . . . . . . . . . . .

. 150

Table G3: Response of international bond yields to monetary shocks

. . . . . . . . . . . . 151

Table G4: Response of international bond yields to monetary and macro data news shocks . 152

Table G5: Response of expected component and term premium component of interna-

tional bond yields to monetary shocks . . . . . . . . . . . . . . . . . . . . . . . 153

Table G6: Response of exchange rates to monetary shocks . . . . . . . . . . . . . . . . . . 154

Table G7: Response of term premium component of international bond yields to monetary

shocks (bond substitutability interaction)

. . . . . . . . . . . . . . . . . . . . . 154

viii

Table G8: Response of US holdings of foreign bonds to monetary shocks . . . . . . . . .

. 155

Table G9: Understanding the cross-country heterogeneity of asset price responses: Emerg-

ing countries

.

.

.

.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

Table G10: Response of international equity indices to monetary shocks . . . . . . . . . . . 157

Table G11: Response of international bond yields accounting for zero lower bound . . . . . 158

Table G12: Response of forecast dispersion to monetary shocks . . . . . . . . . . . . . . . . 159

Table G13: Data coverage for sample countries

. . . . . . . . . . . . . . . . . . . . . . . . 160

Table G14: Response of international bond yields to monetary shocks with QE dummy . . . 161

Table G15: Response of international bond yields to monetary shocks (Pre-crisis sample) . . 162

Table G16: Response of international asset prices to monetary shocks (Post-crisis sample) . . 163

Table G17: Response of US bond term premia to monetary policy shocks

. . . . . . . . . . 164

Table G18: Understanding the cross-country heterogeneity of asset price responses: Ad-

vanced countries

. .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

Table G19: Understanding the cross-country heterogeneity of asset price responses: Emerg-

ing countries with dollar debt exposure

. . . . . . . . . . . . . . . . . . . . . . 166

Table G20: Response of international bond yields to monetary shocks (1-day window) . . . . 167

Table G21: Response of international bond yields to monetary shocks, controlling for

LIBOR-OIS spread .

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

Table G22: Response of international bond yields to monetary shocks, 2 year yield as mps

. 169

Table G23: Response of international bond yields to monetary shocks, 10 year yield as mps . 170

Table G24: Response of expected component and term premium component of interna-

tional bond yields to monetary policy uncertainty (pre- and post-crisis samples) . 171

Table G25: Response of term premium component of international bond yields to mpu

shocks (bond substitutability interaction)

. . . . . . . . . . . . . . . . . . . . . 172

Table G26: Response of US holdings of foreign bonds to monetary shocks (robustness)

. . . 173

ix

Table G27: Response of international bond yields to monetary shocks, controlling for

target and path factor .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. 174

Table G28: Response of international bond yields to monetary shocks (only scheduled

FOMC meetings) .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

.

.

.

.

.

.

.

Table G29: Response of international bond yields to monetary shocks controlling for

information eﬀect .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. 176

Table G30: Response of international bond yields to monetary shocks with country ﬁxed-eﬀects177

x

LIST OF FIGURES

Figure C1: Concentration of US Banking Industry . . . . . . . . . . . . . . . . . . . . .

. 122

Figure C2: Time Series of Financial Variables, by Firm Size . . . . . . . . . . . . . . . . . 123

Figure C3: Within-Period Model Timeline . . . . . . . . . . . . . . . . . . . . . . . . .

. 124

Figure C4:

IRFs of Financial Variables to Positive TFP & Financial Shocks . . . . . . . . . 125

Figure C5:

IRFs of Capital FOC Components to Positive TFP Shock . . . . . . . . . . . . 126

Figure C6: Wald Test for Structural Break in Cyclicality of Small Firm Financing . . . . . . 127

Figure C7: Time Series of Financial & TFP Shocks

. . . . . . . . . . . . . . . . . . . . . 128

Figure C8:

IRFs of Capital FOC Components to Positive Financial Shock . . . . . . . . .

. 129

Figure C9:

IRFs of Financial Variables: Alternative Household Budget Constraint

. . . . . 130

Figure E1: Firm Leverage Plots .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

Figure E2: Eﬀects of an expansionary monetary policy shock . . . . . . . . . . . . . . .

. 146

Figure H1: Monetary policy uncertainty changes (mpu) on FOMC meeting days . . . . . . 178

Figure H2: 10 Year Yield Response on Prominent Monetary Policy Uncertainty Dates . . . 179

Figure H3: Response of 3 month yield to monetary policy uncertainty . . . . . . . . . . .

. 180

Figure H4: Correlation between change in monetary policy uncertainty (mpu) and mon-

etary policy surprise (mps)

. . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

Figure H5: Standard deviation of international asset prices . . . . . . . . . . . . . . . . . . 182

Figure H6: 10 Year Yield Response on Prominent Monetary Policy Uncertainty Dates . . . 183

Figure H7: Persistence of international asset price response to mpu . . . . . . . . . . . . . 184

Figure H8: Heterogeneity in response to mpu across countries . . . . . . . . . . . . . . . . 185

Figure H9: Robustness of capital account openness interaction with mpu . . . . . . . . . . 185

xi

Figure H10: 2 and 10 Year Yield Response to Monetary Policy Uncertainty. Dropping

One Country at a Time.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186

Figure H11: 2 and 10 Year Yield Response to Monetary Policy Uncertainty. Dropping

One FOMC Date at a TIme. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

xii

CHAPTER 1

FINANCIAL CONSOLIDATION AND THE CYCLICALITY OF CORPORATE

FINANCING

1.1 Introduction

The concentration and complexity of ﬁnancial institutions can have major consequences for
ﬁrms’ access to ﬁnance and, ultimately, for investment and production decisions. An aspect that
thus far has received limited attention is the way the structure of the ﬁnancial sector inﬂuences
the behavior of ﬁrms’ ﬁnancing over the business cycle. And yet, studying the drivers of cyclical
ﬁnancing patterns is critical for understanding ﬁrms’ resilience to macroeconomic disturbances and
the mechanisms of propagation of real and ﬁnancial shocks. The goal of this paper is to investigate
how changes in the degree of concentration and complexity of ﬁnancial institutions shape ﬁrms’
ﬁnancing behavior over the business cycle. The United States provides a natural setting for our
analysis.
In the mid-1990s, banking regulatory reforms, especially the Riegle-Neal Interstate
Banking and Branching Eﬃciency Act, allowed for ﬁnancial institutions to engage in acquisitions
and mergers across state lines. Since then, the US ﬁnancial sector has seen a dramatic consolidation,
with ﬁnancial institutions becoming larger and more complex. Figure C1 illustrates the increase in
the concentration of the US banking sector, as measured by the Herﬁndahl-Hirschman indices of
bank loans and assets. In recent years, the share of banking assets held by the top 10 bank-holding
companies has exceeded 60% (Fernholz & Koch (2016)).1

The literature has established three key facts about the cyclical behavior of publicly-traded
ﬁrms’ ﬁnancing.2 First, ﬁrms of all sizes borrow procyclically, that is, during economic expansions
ﬁrms increase their debt.3 Second, small and medium-sized ﬁrms issue equity procyclically, while

*This chapter is joint work with Raoul Minetti and Sotirios Kokas.
1See Berger et al. (1999) for further discussion on the implications of ﬁnancial consolidation.
2Covas & Den Haan (2011), Jermann & Quadrini (2012), Karabarbounis et al. (2014) and Begenau & Salomao (2019)
are primary examples.
3Unless otherwise stated, a “ﬁrm” refers to a publicly-traded US ﬁrm.

1

larger ﬁrms issue equity countercyclically. Third, liquidity accumulation follows the same pattern
as equity issuance, with small and medium-sized ﬁrms increasing cash holdings during economic
expansions and large ﬁrms increasing cash holdings during downturns. In this paper, we conﬁrm
these facts for the period 1981-2017; however, we show that the procyclical equity issuance and
liquidity accumulation of small and medium-sized ﬁrms is driven by the latter half of this period.
From the early 1980s to the late 1990s, in fact, equity ﬁnancing and liquidity accumulation for ﬁrms
of all sizes was countercyclical. After the mid-1990s banking regulatory reforms, the cyclicality
of equity issuance and liquidity accumulation has changed dramatically for the ﬁrms most likely
to have been impacted by the reforms: small and medium-sized ﬁrms with lower bargaining power
vis-à-vis banks and that experienced an increase in the complexity of the relationships with their
lending banks.

Following prior literature, we use data on US ﬁrms from the Compustat North America database
to construct two types of data sets: aggregate time series and ﬁrm-level panel. We then document
the cyclicality of ﬁnancing during the 1981-2017 period for a sample of 16,675 ﬁrms and uncover
a change in ﬁnancing behavior for small and medium-sized ﬁrms in the late 1990s.4 This change
occurred earlier in US states that implemented earlier the Riegle-Neal Act, pointing to a role for the
reforms that led to ﬁnancial consolidation. Next, we match Compustat ﬁrms to syndicated loan data
from Thomson Reuters LPCs DealScan database for the years 1987-2012. Using characteristics
of lenders and loans, we provide further evidence that the US ﬁnancial consolidation contributed
to the change in cyclicality for small and medium-sized ﬁrms. Speciﬁcally, procyclical equity
issuance and liquidity accumulation after the late 1990s is most prevalent among ﬁrms with a
weaker relationship to their lenders, i.e. ﬁrms whose lenders were involved in a large merger or
were acquired by a multi-bank holding company (increased size and complexity of lenders) and
ﬁrms with a smaller set of available lenders (weaker ﬁrm bargaining power vis-à-vis lenders).

After uncovering evidence consistent with US ﬁnancial consolidation as a driver of the changed
cyclicality of small ﬁrm ﬁnancing, we rationalize the empirical ﬁndings though a general equilib-

4In Section 1.4, we perform a Wald test to identify 1999 as the break year.

2

rium business cycle model with ﬁnancial constraints. In the model economy, ﬁrms can borrow
and issue equity to cover short-term and long-term ﬁnancing needs (see, e.g., Jermann & Quadrini
(2012)). Access to bank credit is constrained by imperfect enforcement of debt contracts. Access to
equity markets entails equity issuance costs. We additionally allow ﬁrms to endogenously accumu-
late liquidity and directly bargain with their lending banks over the cost of short-term borrowing.
The threat point of a ﬁrm in these negotiations is increased by holding liquidity, as this allows
the ﬁrm to cover its short-term ﬁnancing needs in case of withholding of bank credit.5 Financial
consolidation is then simulated in the model by weakening ﬁrms’ bargaining power vis-à-vis their
lending banks and strengthening banks’ outside option. This ﬁnancial consolidation produces
cyclical ﬁnancing patterns in line with those documented for small and medium-sized publicly-
traded ﬁrms in the 1999-2017 period. Speciﬁcally, the documented change in cyclicality occurs in
response to TFP shocks in the model, rather than shocks to ﬁrms’ ﬁnancial constraints. We show
that this also holds true empirically.

The intuition for the theoretical results revolves around the idea that, after ﬁnancial consolida-
tion, smaller ﬁrms have stronger incentives to issue equity and accumulate precautionary liquidity
following positive shocks. In particular, a ﬁrm’s demand for labor can be met by accessing short-
term bank credit or drawing down accumulated liquidity.6 The ﬁrm bargains with the lender over
the cost of bank credit. If a ﬁrm has high bargaining power vis-à-vis its lending bank and/or the
bank has less valuable alternatives to lending to the ﬁrm, then the cost of accessing bank credit is
low. When a positive TFP shock occurs, ﬁrms want to increase their labor. A ﬁrm with a low cost
of accessing short-term bank credit (a “large” ﬁrm) simply increases borrowing and pays out higher
proﬁts from the positive shock to equity holders. A ﬁrm with a high cost of accessing bank credit
(a “small” ﬁrm) will also desire to increase its labor; however, the lender can extract a high share of
the surplus of doing so. In response, the ﬁrm will have the incentive to carry more liquidity to oﬀset
the bank’s bargaining advantage. The ﬁrm ﬁnances this extra precautionary liquidity by issuing

5As it realistically takes time for a ﬁrm to issue equity, by assumption the ﬁrm cannot access equity at the time of
bargaining with the lender; thus, the ﬁrm wants to be holding accumulated liquidity.
6See Lins et al. (2010) for evidence that ﬁrms substitute between cash and credit lines.

3

equity. As a result, both liquidity accumulation and equity issuance increase following a positive
TFP shock, i.e.
they behave in a procyclical manner. This pattern does not hold for a positive
ﬁnancial shock, as the loosening of the ﬁrm’s borrowing constraint allows for a ﬁrm (regardless of
bargaining power or the lender’s outside option) to simply increase its debt issuance rather than its
liquidity accumulation and equity issuance.

We next investigate how the changes in cyclical ﬁnancing behavior prompted by ﬁnancial sector
consolidation aﬀect the cyclical behavior of ﬁrms’ investment and employment. Empirically,
the investment and employment of small and medium-sized ﬁrms show an increased sensitivity
to shocks in the post-ﬁnancial sector consolidation period. By contrast, large ﬁrms’ investment
and employment sensitivity does not change following consolidation.
In the model economy,
pronounced eﬀects of shocks on investment occur in the post-ﬁnancial sector consolidation period
via the ﬁnancial channel illlustrated above. Speciﬁcally, ﬁrms’ liquidity holdings magnify their
ability to appropriate surplus when negotiating with banks, as well as increase the value of capital
as collateral. The procyclical liquidity accumulation post-ﬁnancial consolidation, therefore, boosts
ﬁrms’ returns from accumulating capital as a productive input and as collateral, increasing the
sensitivity of investment to shocks.

Related literature. This paper contributes to three strands of literature. The ﬁrst investigates
empirically and theoretically the behavior of ﬁrm ﬁnancing and liquidity accumulation over the
business cycle. Covas & Den Haan (2011), Jermann & Quadrini (2012), Karabarbounis et al. (2014)
and Begenau & Salomao (2019) generally ﬁnd that debt issuance is procyclical in samples that begin
in the early 1980s, while the cyclicality of equity issuance depends on ﬁrm size. Karabarbounis
et al. (2014) show that equity issuance is procyclical for smaller ﬁrms and countercyclical for
large ﬁrms. We conﬁrm these prior studies for the baseline period of 1981-2017; however, we
show that the ﬁnding of procyclical equity issuance for smaller ﬁrms is driven by the post-ﬁnancial
consolidation period. On the theoretical side, Covas & Den Haan (2012) and Jermann & Quadrini
(2012) introduce ﬁnancial frictions to generate a tradeoﬀ between debt and equity ﬁnancing over
the business cycle. Unlike Covas & Den Haan (2012), this paper develops a general equilibrium

4

model, which allows for the incorporation of employment. Further, neither of these models have
a role for ﬁrm liquidity. Bacchetta et al. (2019) introduce ﬁrm liquidity holdings into a general
equilibrium model in which ﬁrms pay wages using external ﬁnancing or internal liquidity. Our
model diﬀers from Bacchetta et al. (2019) by allowing for bargaining between ﬁrms and their
lenders. This allows for an analysis of the eﬀects of ﬁnancial consolidation on the cyclicality of
ﬁrm ﬁnancing.

The second strand of related literature investigates the eﬀects of ﬁnancial sector consolidation on
non-ﬁnancial ﬁrms. Di Patti & Gobbi (2007) show that Italian bank mergers reduced availability of
credit to ﬁrms. Karceski et al. (2005) ﬁnd that bank mergers in Norway lowered the equity value of
publicly-traded ﬁrms that borrow from the merging banks. Carow et al. (2003) show that US bank
mergers have negative equity eﬀects for publicly-traded companies by decreasing their bargaining
power vis-à-vis the merging lenders. We contribute to this literature by exploring the impact of
ﬁnancial consolidation on the cyclicality of ﬁrms’ ﬁnancing, and the consequences to the cyclical
behavior of investment and employment.

Finally, we contribute to the literature on the relative importance of TFP shocks and ﬁnancial
shocks to the business cycle. Jermann & Quadrini (2012) ﬁnd that ﬁnancial shocks are an important
driver of the US business cycle. In contrast, Zetlin-Jones & Shourideh (2017) and Guo (2019)
suggest that ﬁnancial shocks have small eﬀects on real GDP. We ﬁnd that large ﬁrms’ debt issuance,
equity issuance and liquidity accumulation respond to ﬁnancial shocks. This was true of smaller
ﬁrms as well, prior to ﬁnancial consolidation, while equity issuance and liquidity accumulation are
mostly driven by TFP shocks in the post-consolidation period. Through this ﬁnancial channel, TFP
shocks could have gained relative importance over the past two decades.

The remainder of the paper is organized as follows. Section 2 details data and variable
deﬁnitions. Section 3 presents the empirical results using aggregate time series and ﬁrm-level
panel data. Section 4 shows that the empirical results are robust to alternate speciﬁcations.
Section 5 utilizes syndicated loan data to provide empirical evidence that ﬁnancial consolidation
can help explain the empirical ﬁndings. Section 6 describes the business cycle model. Section

5

7 simulates ﬁnancial consolidation in the model. Section 8 provides empirical evidence on the
relative importance of TFP shocks and ﬁnancial shocks to the cyclicality of ﬁnancing behavior.
Section 9 presents evidence that smaller ﬁrms’ investment and employment have become more
sensitive to shocks in the post-ﬁnancial consolidation period. Section 10 concludes. Additional
results are relegated to the Online Appendix.

1.2 Empirical Evidence

This section describes the data sources we use to explain how and why ﬁrm ﬁnancing behavior
has changed over time. This paper uses two primary data sources: Compustat and DealScan.
Compustat provides balance sheet data for publicly-traded ﬁrms. The DealScan database contains
information describing the syndicated lenders for ﬁrms in the Compustat sample. We complement
these primary sources with the Call Report data from the Federal Reserve Bank of Chicago for
information on the bank-holding status of ﬁnancial institutions and the Merger Description data
from the Federal Reserve Bank of Chicago for information on the timing and characteristics of
bank mergers.

1.2.1 Firm-Level Data

The 1981-2017 Compustat North America - Fundamentals Annual ﬁles include publicly-traded
ﬁrms. Compustat ﬁrms account for approximately one fourth of total private sector U.S. employ-
ment; thus, they represent an economically important sample of businesses (see, e.g., Davis et al.
(2006)). We are interested in the eﬀects of consolidation among ﬁnancial institutions (banks), and
publicly-traded ﬁrms may not be as reliant on bank debt as private ﬁrms. However, Table B1 shows
that bank debt accounts for an important share of total debt amongst Compustat ﬁrms: over 20%
of total debt for the average ﬁrm in Compustat.7 Bank debt has also been shown to play a key
role in the sensitivity of Compustat ﬁrms to shocks. For example, Ippolito et al. (2018) ﬁnd that

7We proxy for bank debt by subtracting commercial paper (CMP) from long-term debt - other (DLTO), as in Lee
(2017). Alternatively, Crouzet (2020) creates a bank debt proxy by summing DLTO and notes payable (NP). Using
this alternative measure would result in an even higher bank debt share for Compustat ﬁrms.

6

Compustat ﬁrms with a higher ratio of bank debt to assets are more sensitive to monetary policy
shocks.

The relevant variables for our analysis are primarily those reported in the cash ﬂow statement,
which are not well-populated prior to 1981.8 Firms incorporated outside of the United States are
dropped from the sample. Financial ﬁrms (SIC 6000-6999), utility ﬁrms (SIC 4900-4999) and
quasi-governmental ﬁrms (SIC 9000-9999) are also excluded. The latter two groups are heavily
regulated, which makes their ﬁnancing decisions distinct from other corporate ﬁrms. Similarly,
ﬁnancial ﬁrms are subject to regulations, such as capital requirements, that uniquely aﬀect their
ﬁnancing behavior. As in Covas & Den Haan (2011), three additional restrictions are made. First,
we remove any ﬁrm that engaged in a major merger during the 1981-2017 time period.9 Second, we
remove General Electric, General Motors, Ford and Chrysler, as these ﬁrms were heavily aﬀected
by the FASB94 accounting rule instituted in 1988. Third, we drop any ﬁrm-year observations where
the accounting identity (assets = liabilities + equity) is violated by more than 10% of the ﬁrm’s
book value of assets. Finally, any ﬁrm-year observations with missing values for assets, liabilities,
equity, debt, cash or (net) capital stock are dropped.

Creation of the primary ﬁnancing variables most closely follows Eisfeldt & Muir (2016). Net
debt issuance is computed as long-term debt issuance (DLTIS) minus long-term debt reduction
(DLTR) plus changes in current debt (DLCCH) minus (net) interest paid (XINT).10 Net equity
issuance is the sale of common and preferred stock (SSTK) minus the purchase of common and
preferred stock (PRSTKC) minus cash dividends (DV).11 Liquidity accumulation is deﬁned as the

8Other papers in this literature also tend to begin their samples in the early 1980s, due to changes in U.S. ﬁnancial
markets and the general behavior of numerous economic variables, e.g. the Great Moderation. In the Online Appendix,
we show that the results are virtually unchanged if we start the sample in 1984, rather than 1981.
9A “major merger” is deﬁned as any merger or acquisition where sales increased by at least 50 percent afterwards
(Compustat sales footnote code AB).
10For ﬁrms with an scf code of 1, DLCCH is subtracted. As described in Chang et al. (2014), prior to the adoption of
uniform reporting rules in 1988, DLCCH that was reported on a ﬁrm’s working capital statement (scf = 1) has the
opposite sign as when reported on other ﬁnancial statements. For ﬁrms with a data code of 4, DLCCH was assumed
to be included in DLTIS, so DLCCH was set to zero.

11Missing values of DV are set to zero in order to avoid too many missing values for net equity issuance. We show in
the Robustness section that using net sale of stock (i.e. SSTK minus PRSTKC) produces results very similar to the
equity issuance measure. Thus, the results are not driven by the behavior of dividends.

7

change in cash and cash equivalents (CHEt - CHEt−1).12 All variables are normalized by the
lagged book value of total assets (AT). We show in the Robustness section that the results hold if
we instead normalize by the lagged (net) capital stock (PPENT).

As in Covas & Den Haan (2011), ﬁrms are grouped into size bins using acyclical cutoﬀs of the
book value of total assets. Speciﬁcally, ﬁrms are ﬁrst split into size groups by the previous year’s
asset value. We deﬁne small ﬁrms as those with a book value of assets below the 60th percentile
and large ﬁrms as those above the 60th percentile (excluding the top 10 percent of ﬁrms).13 A (log)
linear trend is then ﬁt through these annual cutoﬀ values and used as the new cutoﬀ values for ﬁrm
size groupings. This prevents the cutoﬀ values themselves from being cyclical; however, the results
using the original cutoﬀ values are very similar to those with the adjusted values.

For the aggregate time series results, we follow a similar methodology as Eisfeldt & Muir
(2016) and Covas & Den Haan (2011). Speciﬁcally, we sum the ﬁnancing variable of interest
for all ﬁrms of a size classiﬁcation within a year. Then, we divide each series by the sum of the
asset value for all ﬁrms of a size classiﬁcation within a year to create the aggregate series by size.
Finally, we HP ﬁlter the aggregate ﬁnancing series to produce a stationary series.14 The cyclical
component of this HP-ﬁltered series is then used in all correlations to remove the longer-run trends
in the variables. While it has been standard to use HP ﬁltering in this literature, Hamilton (2018)
warns that HP ﬁltering can cause spurious correlations. We show in the Robustness section that
the results hold if we use either the non-HP-ﬁltered ﬁnancing series or if we ﬁlter based on the
Hamilton (2018) methodology.15 Further, the ﬁrm-level panel regressions do not use ﬁltering and
still produce results similar to those found with the HP-ﬁltered aggregate series; thus, the results
are not driven by the choice to use HP ﬁltering.

12We use the balance sheet version of cash, rather than the cash ﬂow statement version (CHECH). This is due to
CHECH being unavailable prior to 1984. We show in the Robustness section that the results hold using either version
of liquidity accumulation, as well as using change in cash only (CHt - CHt−1), i.e. excluding cash equivalents.
13See Eisfeldt & Muir (2016) for a description of how the top 10 percent of ﬁrms present measurement problems and
anomalous ﬁnancing behavior that makes their inclusion in the sample misleading of ﬁrm dynamics.

14For the baseline results, we use annual data; thus, the smoothing parameter is set to 100. For quarterly data, the

smoothing parameter is set to 1600.
15To perform the Hamilton (2018) ﬁltering, we use the Diallo (2018) HAMILTONFILTER Stata command. For annual
data, this amounts to using the residual from regressing the variable of interest at year t on its values at year t − 2 and
t − 3.

8

Splitting by ﬁrm size categories, Table B1 shows summary statistics of asset value, ﬁrm age and
key ﬁnancing variables for the 16,675 ﬁrms in the sample. Since all ﬁrms are publicly-traded, even
“small ﬁrms” are quite large relative to the typical private ﬁrm. Still, there is a sizable discrepancy
between the ﬁrm categories:
the average small ﬁrm has an asset value of $71.5 million, while
the average large ﬁrm has an asset value of $931 million. As expected, larger ﬁrms tend to be
older. Smaller ﬁrms rely more on equity ﬁnancing than larger ﬁrms and also tend to accumulate
more liquidity. During the sample period 1981-2017, approximately 90% of ﬁrms fall within
their modal ﬁrm size category. Put diﬀerently, ﬁrms rarely cross size bins. This suggests we can
(approximately) treat ﬁrm size as a ﬁxed ﬁrm characteristic.

1.2.2 Lender Data

We use information on syndicated loans from the Thomson Reuters LPC’s DealScan database for
the years 1987-2012. This database allows us to link syndicated lenders to their borrowing ﬁrms
in Compustat. The syndicated loan market consists of groups of lenders that jointly loan funds
to a single borrowing ﬁrm. A subset of the lenders in a syndicate are the lead arrangers. The
lead arrangers agree with the ﬁrm on the key characteristics of the loan, including loan amount,
collateral, and interest rate. They are also responsible for inviting the other syndicate lenders
to join. The non-lead members of the syndicate (“participants”) provide funds and assist in the
administrative tasks (Delis et al. (2017)).

Using the DealScan database, we create a pool of lenders for each Compustat ﬁrm that matches
to DealScan. Speciﬁcally, any lender that was engaged in a syndicated loan relationship with a ﬁrm
in the current year, the previous 5 years or the next 5 years are classiﬁed as belonging to a ﬁrm’s
lender pool. Since ﬁrms do not necessarily participate in the syndicated loan market every year,
using a window of ±5 years allows us to capture those banks that act as key lenders to the ﬁrm in
the current period.16 Both the lead lenders and participants interact, and contract, directly with the
ﬁrm (Mugasha (1998)). This allows the lenders, including the participant lenders, to gain important

16The results are generally robust to using a diﬀerent window length.

9

information about the borrowing ﬁrm through the syndicated loan agreement (Li (2017)).

After creating this lender pool, we are interested in measures of the relative power of the lenders
vis-à-vis the borrowing ﬁrms. We use various proxies: the total number of lenders in a ﬁrm’s pool,
the share of syndicated loans provided by the lead lender(s), an indicator for a lender recently being
acquired by another lender, and an indicator for a lender recently joining a multi-bank holding
company. The last two indicators represent exogenous changes in the relationship between a ﬁrm
and a lender.
In Section 1.5, we use these proxies to present evidence that an increase in the
size and complexity of the lending banks is a key driver in changing the cyclicality of small and
medium-sized ﬁrms’ ﬁnancing behavior.

1.3 Firm Financing over the Business Cycle

In this section, we present evidence of a structural break in the ﬁnancing behavior of small and
medium-sized publicly-traded ﬁrms in the late 1990s. In an April 2001 speech, Federal Reserve
Vice Chairman Roger Ferguson noted that “Financial consolidation has helped to create a signiﬁcant
number of large, and in some cases increasingly complex, ﬁnancial institutions” and that “the pace
of consolidation increased over time, including a noticeable acceleration in the last three years
of the [1990s]” (Ferguson (2001)). A key contributor to this consolidation was the Riegle-Neal
Interstate Banking and Branching Eﬃciency Act of 1994, which applied to states at diﬀerent dates
between 1994 and 1997.17 Heiney (2010) presents evidence that the banking sector consolidation
of the 1990s begins to slow down after 1999. As shown in Figure C1, the concentration of the
banking sector, as measured by Herﬁndahl-Hirschman indices of bank loans and assets, noticeably
increases between the passage of Riegle-Neal in 1994 and the end of the 1990s. The other major
ﬁnancial reform of the 1990s, the Gramm-Leach-Bliley Act, was passed in 1999 and allowed for
bank holding companies to integrate their commercial banking activities with investment banking.
For the above reasons, the year 1999 serves as a natural partition for our analysis.18

We show that, prior to 1999, the cyclicality of debt issuance, equity issuance and liquidity

17See Dick (2006) for a list of state-speciﬁc dates of adoption.
18Additional empirical evidence for using the year 1999 is presented below in the Robustness section.

10

accumulation behaves similarly for smaller and larger ﬁrms. But, after 1999, the ﬁnancing behavior
of small and medium-sized ﬁrms becomes signiﬁcantly diﬀerent than larger ﬁrms. Speciﬁcally,
these ﬁrms begin to issue equity and accumulate liquidity in a procyclical fashion. These results
hold using both the correlations of aggregated time series data and panel regression estimates of
disaggregated ﬁrm-level data. The following sections explore the potential causes of this change in
cyclicality and ﬁnd that the mechanisms are consistent with ﬁnancial consolidation: a weakening
of smaller ﬁrms’ bargaining power vis-à-vis their lending banks and an increase in the complexity
of their relationships with banks.

1.3.1 Aggregate Time Series Evidence

To visually illustrate the main stylized facts documented in this paper, Figure C2 plots the aggregate
time series of debt issuance (Panel a) and equity issuance (Panel b) separately for “small” (asset value
below the 60th percentile) and “large” ﬁrms (asset value between the 60th and 90th percentiles),
as well as the cyclical component of real corporate GDP. All series are standardized to have a
mean of zero and unit variance. In Panel (a), the debt issuance series for small and large ﬁrms
essentially overlap each other for the entire 1980-2017 period. They also clearly comove positively
with cyclical GDP, i.e. debt issuance is procyclical for both small and large ﬁrms. In Panel (b),
the equity issuance series for both small and large ﬁrms positively comove throughout the ﬁrst
half of the sample period; however, these series negatively comove during the 2000s. In terms of
cyclicality, both small and large ﬁrms negatively comove with cyclical GDP in the ﬁrst half of the
time period, while the equity issuance of small ﬁrms positively comoves in the latter half. The
behavior of the liquidity accumulation series is quite similar to that of equity issuance. In what
follows we more rigorously demonstrate that the cyclicality of small ﬁrms’ equity issuance and
liquidity accumulation changed in the late 1990s and uncover a role for ﬁnancial consolidation in
this phenomenon.

Using the aggregate series displayed in Figure C2, Table B2 presents the correlation between

11

each of the main ﬁnancing variables and the cyclical component of real corporate GDP.19 Panel
A shows the correlations for the entire 1981-2017 period.20 As commonly found in the literature,
debt issuance is strongly procyclical for ﬁrms of all sizes, while equity issuance and liquidity accu-
mulation are procyclical for smaller ﬁrms and countercyclical for large ﬁrms. One of the insights of
this paper is evident by comparing Panel B to Panel C: the commonly found procyclicality of small
ﬁrms’ equity issuance and liquidity accumulation is driven by the post-1999 time period (Panel C).
One can see from Panel B that equity issuance and liquidity accumulation are countercyclical for
smaller ﬁrms prior to 1999. In contrast, large ﬁrms behave virtually the same in both periods.

1.3.2 Firm-Level Evidence

Next, we reproduce these cyclical ﬁnancing patterns using ﬁrm-level panel data, which allows for
the addition of ﬁrm characteristics beyond size. Table B3 displays results from estimating a total of
24 regressions with the following 3 speciﬁcations. First, the reported coeﬃcients are estimated by
12 regressions (2 time periods x 3 ﬁnancing variables x 2 size groups) of the following speciﬁcation:

Vi,t = α0 + α1t + α2t2 + βYt + Γ(cid:48)Zi,t−1 + ei,t

(1.1)

where Vi,t is the ﬁnancing variable of interest normalized by the lagged book value of assets, α0 is
a constant 21, t and t2 capture trends in the ﬁnancing variable, Yt is the cyclical component of real
corporate GDP normalized such that a unit increase in Yt indicates moving from the lowest value
of the cyclical component to the highest value during the sample period 1981-2017, Zi,t−1 includes
the lagged values of the controls, and ei,t is the error term. For the baseline speciﬁcation, we follow
Covas & Den Haan (2011) and include the following controls: ﬁrm’s cash ﬂow and Tobin’s Q,
where each control variable is the diﬀerence between the ﬁrm’s value at t −1 and the respective size
group’s mean value at t − 1. This normalization prevents the controls from picking up variations

19Following most of the literature, we use the cyclical component of HP-ﬁltered real corporate GDP to measure the

business cycle. In the Robustness section, we show that the results are robust to alternative measures.

20Since we HP ﬁlter the variables for the baseline results, there is potential for end-point bias. In the Online Appendix,

we drop the ﬁrst and last three years from the baseline sample and show that the results are nearly identical.

21Next we show that including a ﬁrm ﬁxed eﬀect produces similar results.

12

in aggregate economic conditions.22 We report the β coeﬃcient in the tables, with standard errors
in parentheses clustered along both the time and ﬁrm dimensions.

Second, the reported p-values are the result of 6 regressions (2 time periods x 3 ﬁnancing

variables) where the 2 ﬁrm size groups are pooled:

Vi,t = αj + I(j)i,t(α1, jt + α2, jt2 + βjYt + Γ(cid:48)

j Zi,t−1) + ei,t

(1.2)

where αj is a size group j ﬁxed eﬀect and I(j)i,t is an indicator for the size group to which ﬁrm i
belongs to in year t. This indicator is interacted with all of the explanatory variables. We report the
p-values of βlarge, where small is the base group. These p-values indicate whether the cyclicality
of the small group statistically diﬀers from the large group.

The bold coeﬃcients in the post-1999 period are based on the p-values from 6 regressions (3

ﬁnancing variables x 2 size groups) where the 2 time periods are pooled:

Vi,t = αk + I(k)i,t(α1,kt + α2,kt2 + βkYt + Γ(cid:48)

k Zi,t−1) + ei,t

(1.3)

where αk is a time period k ﬁxed eﬀect and I(k)i,t is an indicator for the time period to which ﬁrm
i belongs to in year t. This indicator is interacted with all of the explanatory variables. We bold the
coeﬃcients in the post-1999 period to indicate a p-value below 0.05 for the coeﬃcient βpost1999,
where pre1999 is the base group. These p-values indicate whether the cyclicality of the variable
of interest is statistically diﬀerent in the 1999-2017 period, relative to the 1981-1998 period.

Panel A of Table B3 reports the baseline panel results. The coeﬃcients can be interpreted
as the eﬀect on the ﬁnancing variable (as a percentage of the ﬁrm’s asset value) of moving from
the lowest realization of the business cycle measure in the full sample period 1981-2017 to the
highest realization, i.e. a positive coeﬃcient indicates a procyclical relationship. First, note that
the strength and direction of the cyclicality implied by the coeﬃcients in Panel A qualitatively
match quite closely with the aggregate correlations in Panel B and Panel C of Table B2. Thus,

22Using a small number of controls allows for a parsimonious model; however, it does not rule out the possibility that
the estimated cyclicality by ﬁrm size is driven by omitted non-size variables. On the one hand, ﬁrm size is intended
to broadly capture characteristics shared by ﬁrms of similar size. Still, in the Online Appendix we show that the ﬁrm
size results hold when we allow a wider set of ﬁrm characteristics to explain cyclicality in ﬁnancing variables.

13

while the panel regressions treat each ﬁrm observation equally, the results are quite similar to the
aggregate data, where ﬁrms are weighted by their asset value. Second, the p-values in the pre-1999
period for equity issuance and liquidity accumulation are at or below 0.10, indicating statistically
signiﬁcant diﬀerences in the cyclicality of equity issuance between small ﬁrms and large ﬁrms.
Third, as indicated by the bold coeﬃcients in Panel A, the cyclicality of equity issuance and
liquidity accumulation for small ﬁrms are signiﬁcantly diﬀerent in the post-1999 period than the
cyclicality of small ﬁrms in the pre-1999 period. For small ﬁrms, the sign of the coeﬃcient ﬂips,
with the magnitude of the coeﬃcients in the post-1999 period being quite similar to the pre-1999
coeﬃcients in absolute value. In sum, we detect large and statistically signiﬁcant changes in the
ﬁnancing behavior of only small ﬁrms for the post-1999 period.

We can get a sense of the magnitude of changes in GDP on equity issuance by converting to a
one-standard deviation change in cyclical GDP. Moving from the lowest realization of the business
cycle measure in the full sample period 1981-2017 to the highest realization is approximately a 4.5
standard deviation change. Thus, dividing the post-1999 equity issuance coeﬃcient for small ﬁrms
of 11.12 by 4.5 results in a standardized coeﬃcient of 2.5. Comparing to the simple average equity
issuance for small ﬁrms of 19.1% of assets (see Table B1), this amounts to an eﬀect equivalent to
13% of average equity issuance. Alternatively, we could compare to the average annual aggregate
equity issuance for small ﬁrms of 9.85%. This amounts to an eﬀect equivalent to 25% of average
equity issuance. Changes in GDP thus appear to have an economically signiﬁcant eﬀect on ﬁrm
equity issuance.

To this point, the regressions have not included a ﬁrm ﬁxed eﬀect; however, it could be the case
that an idiosyncratic ﬁrm component is responsible for the results. We next include a ﬁrm ﬁxed
eﬀect to control for permanent heterogeneity. Since the Compustat sample contains a substantial
amount of ﬁrm entry and exit, we do not want the ﬁrm ﬁxed eﬀect to be endogenous. As a result, we
keep only ﬁrms that have greater than 5 years of data within a subperiod. In Table B11, we report
the results of estimating the baseline speciﬁcation while including this ﬁrm-speciﬁc constant term.
As with the baseline speciﬁcation, equity issuance and liquidity accumulation are countercyclical

14

in the pre-1999 period and signiﬁcantly more procyclical in the post-1999 period.

However, since ﬁrms rarely move between the two ﬁrm size categories, the inclusion of a ﬁrm
ﬁxed eﬀect does not exploit within-ﬁrm size variance over time. As an alternative, we next use a
continuous ﬁrm size measure. Speciﬁcally, we want to answer the question, “During the post-1999
period, does equity issuance and liquidity accumulation for a ﬁrm become less procyclical as that
individual ﬁrm grows?”We re-estimate the baseline panel speciﬁcation with the following changes:
the addition of a ﬁrm ﬁxed eﬀect, the use of a continuous measure of size (i.e. log of asset value) and
the demeaning of the continuous size measure at the ﬁrm level. The speciﬁcation is the following:23

Vi,t = αi + α1t + α2t2 + β1(sit − Ei[si])Yt + Γ(cid:48)Zi,t−1 + ei,t

(1.4)

where αi is a ﬁrm i ﬁxed eﬀect. The variable sit measures ﬁrm i’s size (i.e. log of asset value) in
year t and Ei[si] is the average size of ﬁrm i in the sub-period. With this speciﬁcation, the cyclicality
of a ﬁrm’s ﬁnancing variables is identiﬁed by the variation in a ﬁrm’s current size relative to that
ﬁrm’s average size.

In the post-1999 period, one would anticipate that a ﬁrm’s equity issuance and liquidity accu-
mulation becomes signiﬁcantly less procyclical as it grows, i.e. looks more like a “large” ﬁrm. The
results in Panel B of Table B3 show that this is true.24 As well, this was clearly not true of the
pre-1999 period. Given these empirical results, Section 5 will use information about ﬁrms’ lenders
to investigate what caused smaller ﬁrms’ ﬁnancing behavior to change post-1999.

1.4 Robustness

This section ﬁrst presents evidence that the change in smaller ﬁrm behavior occurred around
1999, consistent with the importance of ﬁnancial consolidation. Then, we show that the empirical
results are robust to alternative speciﬁcations and assumptions. Speciﬁcally, we test the robustness
of the main ﬁnding: equity issuance and liquidity accumulation became procyclical in the 1999-
2017 period for smaller ﬁrms only.

23We again drop ﬁrms that appear in fewer than 6 sample years to prevent the ﬁrm ﬁxed eﬀect from being endogenous.
24The Online Appendix presents the results of using continuous size without a ﬁrm ﬁxed eﬀect or isolating within-ﬁrm

variance. The results are consistent with the baseline categorical size ﬁndings.

15

1.4.1

Identifying the Structural Break

Thus far, we have used 1999 as the year where small ﬁrms experienced a structural break in their
equity issuance and liquidity accumulation. Intuitively, 1999 is a natural break year, as a substantial
portion of the post-reform ﬁnancial consolidation had occurred by 1999. To provide evidence of
the suitability of this assumption, we regress the aggregate ﬁnancing variable series for small ﬁrms
on the cyclical component of real corporate GDP. We then perform a Wald test for a structural
break in the estimated cyclicality coeﬃcient. In Figure C6, we plot the p-values of these Wald
tests for each year from 1984-2014. Unsurprisingly, the p-values for debt issuance are never below
0.1, as there is not strong evidence for a change in the cyclicality of debt issuance. Conversely, the
p-value for a structural break in equity issuance is at its lowest, below 0.1, in 1999.25 The results
for liquidity accumulation are similar, with the p-value falling below 0.1 in 1999. We conclude that
1999 is the strongest candidate for a structural break in the cyclicality estimates.26

1.4.2 Other Robustness Tests

One potential concern with the empirical results is that, rather than capturing the cyclicality of
smaller ﬁrms, they could be capturing the cyclicality of ﬁrm entry.
In other words, during an
expansion, many young ﬁrms choose to go public and disproportionately issue equity. To account
for this, we restrict the sample to only those ﬁrms that entered the Compustat sample prior to 1990
and were also in the sample in 2017. Despite severely restricting the sample, Table B11 shows that
the results hold. Thus, the ﬁndings are not driven by ﬁrm entry or a change in the composition of
the sample over time.

Next, one may prefer alternative deﬁnitions of the ﬁnancing variables. Table B12 displays the
results of alternative deﬁnitions for equity issuance and liquidity accumulation. We split the net
equity issuance variable into its two components: net sale of stock and dividend payouts. Recall

25Similarly, we also rerun the baseline panel results using diﬀerent break years. This method produces similar results
as the Wald test, which is that the pre-1999 vs post-1999 break is the break year with the highest signiﬁcance and the
results weaken as the break year moves away from 1999.

26Given the literature discussed above that shows an increase in ﬁnancial consolidation beginning in the mid-1980s, it

is not surprising that there is a decreasing trend in the p-values of our Wald tests throughout much of the 1990s.

16

that net equity issuance is the sale of common and preferred stock (SSTK) minus the purchase of
common and preferred stock (PRSTKC) minus cash dividends (DV). Table B12 shows that the net
sale of stock (i.e. excluding cash dividends) results closely match the equity issuance results. Thus,
the behavior of dividend payouts does not drive the ﬁndings. The main results also qualitatively
hold for three alternative deﬁnitions of liquidity accumulation. The ﬁrst alternative deﬁnition uses
the cash ﬂow statement version of change in cash and cash equivalents, rather than the balance
sheet version used in the baseline estimates.27 The second alternative deﬁnition uses changes in
cash only, rather than cash and cash equivalents. The third deﬁnition uses retained earnings.

While it is standard to use HP-ﬁltering in this literature, it could be the case that the ﬁltering of
the GDP measure and/or ﬁnancing variables (aggregate results only) are non-trivially impacting the
results. In Table B13 we reproduce the baseline correlation results with the non-ﬁltered ﬁnancing
series and the annual growth rate in real corporate GDP. The main ﬁndings are less strong, but the
general pattern clearly holds. Next, we additionally show that ﬁltering the ﬁnancing series and/or
GDP using the Hamilton (2018) methodology produces qualitatively similar results to the baseline
estimations. Appendix Table B13 shows this for the aggregate correlation results. Overall, the
results are not dependent on the choice to HP-ﬁlter the data.

Finally, we include a few additional robustness checks in the Online Appendix. First, we verify
that the ﬁndings are not driven by the cyclicality or potential endogeneity of asset value. We do
so with two robustness checks: normalizing the ﬁnancing variables by the (net) capital stock value
or by the ﬁrm’s ﬁrst reported asset value. Both normalizations produce results quite similar to our
baseline ﬁndings. Second, we exclude all observations with any merger, rather than just ﬁrms that
experienced a major merger. The results are largely unchanged to this exclusion; thus, the ﬁndings
are not due to, for example, the issuance of equity during the merging process. Third, one might
be concerned that small ﬁrms are more likely to be on the verge of bankruptcy, i.e. ﬁnancially
distressed, and that our results might then be picking up the impact of ﬁnancial distress. We have
already conﬁrmed that the results are not driven by ﬁrm entry or exit. But, in the Online Appendix,

27The cash ﬂow statement version is not available until 1984.

17

we exclude ﬁrms that are considered “ﬁnancially distressed” according to the Altman Z-Score. The
results remain largely unaltered.

1.5 The Role of Financial Consolidation

We study the role of ﬁnancial consolidation in driving the change in small ﬁrms’ ﬁnancing
cyclicality. We ﬁrst exploit the staggered adoption of the Riegle-Neal Interstate Banking and
Branching Eﬃciency Act and show that the change in ﬁnancing cyclicality is most pronounced
amongst ﬁrms headquartered in states that adopted Riegle-Neal earlier. Next, we create a pool of
lenders with whom Compustat ﬁrms have a syndicated loan relationship, as detailed in Section
1.2. The characteristics of these ﬁrms’ key lenders are then used to test whether a change in
bargaining power and bank complexity contributed to the ﬂip in ﬁnancing cyclicality for smaller
ﬁrms beginning in the late 1990s.

1.5.1 Timing of Riegle-Neal Adoption

The Riegle-Neal Interstate Banking and Branching Eﬃciency Act was passed in 1994, but states
individually passed legislation that determined when Riegle-Neal went into eﬀect. This led to
staggered adoption during the years 1995 (14 states), 1996 (12 states) and 1997 (24 states). We test
whether the year of adoption is associated with diﬀerent procyclicality of equity and liquidity for
smaller ﬁrms using the following speciﬁcation:

Vi,t = αh + α1t + α2t2 + I(h)i βhYt + Γ(cid:48)Zi,t−1 + ei,t

(1.5)
where h is the year of Riegle-Neal adoption for the state in which a ﬁrm is headquartered and I(h)i
is an indicator for this year. In Table B4, we display the estimates of β1995, β1996 and β1997 for
smaller ﬁrms in the periods 1981-1998, 1999-2009 and 1999-2019. Panel C (1999-2019) shows
that equity and liquidity are signiﬁcantly less procyclical for ﬁrms headquartered in 1997 adopters.
This is especially true for the ﬁrst decade after reform, as seen in Panel B (1999-2009). These same
trends were not apparent prior to Riegle-Neal (see Panel A: 1981-1998). These results suggest
that the state-speciﬁc adoption of Riegle-Neal noticeably contributed to the change in cyclicality

18

patterns identiﬁed above. In what follows, we use matched bank-ﬁrm information to further test
the eﬀects of ﬁnancial consolidation on ﬁnancing cyclicality.

1.5.2 Measures of Financial Consolidation

a. Reduction in Firms’ Bargaining Power vis-à-vis Lenders

In Table B5, we re-estimate the baseline panel regression with an additional interaction term to test
whether a reduction in ﬁrms’ bargaining power vis-à-vis banks is associated with the documented
change in cyclicality for smaller ﬁrms post-1999:

Vi,t = α0 + α1t + α2t2 + β1Yt + β2Xi,t + β3Yt ∗ Xi,t + Γ(cid:48)Zi,t−1 + ei,t

(1.6)

where Yt ∗ Xi,t is the interaction of our business cycle measure, Yt, with a characteristic of the ﬁrm’s
lender pool, Xi,t. The main coeﬃcient of interest is β3, which captures the eﬀect on the cyclicality
measure, Yt, of moving from the 25th percentile value for the characteristic of the ﬁrm’s lender
pool to the 75th percentile value. Since we are interested in only the interaction term, we could
alternatively replace the trend variables with a year ﬁxed eﬀect to control for omitted aggregate
variables. Doing so does not meaningfully change our estimates of β3.

Sharpe (1990), Rajan (1992), Boot (2000) and Ongena & Smith (2001) show that a single
bank can extract monopoly rents through future loans to the ﬁrm. Borrowing from multiple banks
moderated such “hold-up” issues. In Panel A of Table B5, we use the total number of lenders in
the created ﬁrm’s lender pool as the proxy for a ﬁrm’s outside options in bargaining with a lender.
Here, a higher number of lenders for smaller ﬁrms in the post-1999 period is associated with a less
procyclical equity issuance and liquidity accumulation, as well as a more procyclical debt issuance.
This suggests that the cyclicality of smaller ﬁrms ﬁnancing behavior moves in the direction of larger
ﬁrms cyclicality when smaller ﬁrms have a larger set of lenders. By contrast, no such evidence
emerges for the pre-1999 period.

Next, we proxy for the ﬁrm’s outside option with the lead lender(s) average share of the total
syndicated loan value for a ﬁrm. As demonstrated by Rajan (1992), the larger the share of the lead

19

lender, the stronger the informational monopoly power of the lender vis-à-vis the ﬁrm. Thus, a high
lead lender share suggests that the ﬁrm is more reliant on the lead lender for ﬁnancing. As expected,
in Panel B, we see that the more concentrated the syndicated loans amongst the lead lender(s), the
less that small ﬁrms behave like large ﬁrms, in terms of the cyclicality of debt issuance, equity
issuance and liquidity accumulation.

Importantly, the interaction coeﬃcients for the pre-1999 period are generally insigniﬁcant and
opposite of the post-1999 sign. For example, decreasing the number of lenders in post-1999 is
signiﬁcantly related with a more procyclical equity issuance for smaller ﬁrms; however, there is
no relationship pre-1999. Table B1 shows means for key characteristics of the syndicated lenders.
Our proxies have not seen substantial change from the pre to post periods. But the size/strength of
the lenders in the lender pools has seen a noticeable increase, as illustrated by the average lender’s
share of state assets and the average lender’s Lerner index.28 Thus, for a given reduction in the
number of available lenders for a ﬁrm, the eﬀect is stronger in the post-1999 period. This suggests
that ﬁnancial consolidation inﬂuenced the impact of ﬁrm-bank relationships on ﬁnancing behavior
by increasing the intensity of the eﬀect, i.e. a change in available lenders matters more because the
lenders themselves are “stronger”.

To probe this point further, we next use a proxy for the lender’s market power. We re-estimate
similar regressions as above; however, the interaction term now ﬂags when a ﬁrm’s lender has
recently been acquired by another lender. This plausibly captures an exogenous change in the
market power of a ﬁrm’s lender. Panel A of Table B6 shows evidence for the full 1985-2012
period that a ﬁrm’s lending bank being acquired by another bank matters, i.e.
it is associated
with increased procyclicality of smaller ﬁrms’ equity and liquidity. While we do not ﬁnd that this
marginal eﬀect of mergers is stronger in the post-1999 period, it is the case that mergers occur
twice as frequently in the DealScan sample during this latter period (see Table B1). This suggests
the overall contribution of mergers to smaller ﬁrms’ cyclicality has become more important via the
increased incidence of mergers. Additionally, one would expect that a larger merger would have a

28The Lerner index is the diﬀerence between the price of bank production and the marginal cost, divided by the

marginal cost.

20

greater impact on the borrowing ﬁrms. Panel B of Table B6 presents a triple interaction between
the size of the merger (i.e. the percentage increase in the original lender’s asset value due to the
merger), the incidence of a merger and cyclical GDP. Increasing the size of the merger in the All
Lender Pool by one standard deviation leads to an additional 23.38 percentage point increase in
the procyclicality of equity issuance and an additional 10.56 percentage point increase for liquidity
accumulation.

We can further assess the impact of a bank merger by splitting ﬁrms into those with few outside
options and those with many outside options. One would expect the impact of a bank acquisition
to be greater for those ﬁrms with fewer lenders in their lender pool. In Panel C of Table B6, we
split our sample into ﬁrms with a below average number of lenders (“Few Lenders”) and ﬁrms with
an average or above number of lenders (“Many Lenders”). As expected, ﬁrms with few lenders
whose lead lender was recently acquired by another lender have signiﬁcantly more procyclical
equity issuance. In contrast, ﬁrms with many lenders see no eﬀect from a lead lender acquisition.
Table B14 shows qualitatively similar results for the acquisition of any lender.

b. Increased Complexity in Bank-Firm Relationships

Panel A of Table B7 repeats the same exercise with an interaction term to test whether increased
distance between the ﬁrm and the lender can explain the documented changes in cyclicality. Here,
the term “distance” refers both to physical distance as well as the level of complexity in the
relationship between a ﬁrm and lender. First, we interact the business cycle measure with whether
any of the ﬁrm’s lenders has joined a multi-bank holding company (MBHC) within the past 5 years.
Joining a MBHC is evidence that more of the lender’s decisions are moved away from the local
loan oﬃcers to far-away headquarters. Thus, the bank is less interested in (has a looser link with)
the local ﬁrm.29 Equity issuance is signiﬁcantly more procyclical for those small ﬁrms who have
a lender in their pool who has recently joined a MBHC. There is also a similar eﬀect from a lead
lender joining a MBHC.

Note that the base group, i.e. those ﬁrms without a lender who has recently joined a MBHC,

29Berger et al. (2005) show that larger banks tend to have shorter, more impersonal lending relationships with ﬁrms.

21

still includes many lenders who had previously joined a MBHC more than ﬁve years ago. Thus,
this would be expected to attenuate our estimate of β3. Still, we ﬁnd a signiﬁcant eﬀect consistent
with a weakening in the relationship between a ﬁrm and a lender leading to a more procyclical
equity issuance.

Finally, we can further assess the impact of increased complexity by splitting ﬁrms into those
with few outside options and those with many outside options. In Panel B of Table B7, we split
our sample into ﬁrms with a below average number of lenders (“Few Lenders”) and ﬁrms with an
average or above number of lenders (“Many Lenders”). Firms with few lenders whose lead lender
recently joined a MBHC have signiﬁcantly more procyclical equity issuance. In contrast, ﬁrms
with many lenders see no eﬀect. Table B14 shows qualitatively similar results for any of the ﬁrm’s
lenders joining a MBHC.

1.6 The Model

Motivated by the empirical ﬁndings, we study a general equilibrium business cycle model with
ﬁnancially constrained ﬁrms. In line with the empirical setting, ﬁrms can ﬁnance short-term and
long-term ﬁnancing needs via borrowing or equity issuance. The model economy builds upon
Covas & Den Haan (2012), Hennessy & Whited (2005) and Jermann & Quadrini (2012). Our
model most closely resembles Jermann & Quadrini (2012); however, we depart in at least two key
ways. First, we allow for ﬁrms to hold liquidity. This is important to investigating the comovement
of debt issuance, equity issuance and liquidity accumulation. Second, we posit that ﬁrms bargain
with banks over the cost of loans. This endogenizes the desire for ﬁrms to hold liquidity.

1.6.1 Firm Technology and Financing

Time is discrete and inﬁnite (see Figure C3 for the within-period timeline of the economy). There
is a [0, 1] continuum of ﬁrms with a production function F(zt, kt, nt) = zt k θ
, where zt is
stochastic aggregate productivity, kt is the ﬁrm’s capital stock, and nt is the ﬁrm’s labor. Capital
evolves according to kt+1 = (1 − δ)kt + it, where it is investment and δ denotes the depreciation

t n1−θ

t

22

rate.

Firms have access to three forms of external ﬁnancing: equity issuance, intertemporal debt
(bonds) and an intraperiod loan, obtained from a lender (bank). The amount of intraperiod
borrowing from a bank is subject to constraints, due to enforcement problems. Firms can issue
equity by decreasing their equity payout, dt, where a negative value indicates net equity issuance.
Firms that deviate from the long-run equity payout target are subject to a quadratic cost that makes
the total cost of equity payouts ϕ(dt) = dt + κ · (dt − ¯d)2, where κ ≥ 0 represents the friction
of substituting from debt ﬁnancing to equity ﬁnancing and ¯d is the long-run (steady-state) equity
payout target. Intertemporal debt, bt, has a tax advantage that makes it preferable to issuing equity.
This preference of debt to equity follows the standard pecking order assumption found in models
such as Jermann & Quadrini (2012) and Hennessy & Whited (2005). Speciﬁcally, ﬁrms face an
eﬀective gross interest rate of Rt = 1 + rt(1 − τ), where rt is the interest rate and τ is a tax subsidy.
The intraperiod bank loan will be discussed in further detail below.

Firms can accumulate liquidity, at, and carry it between periods. Holding liquidity allows ﬁrms
to cover current period operating costs and reduces the amount of external ﬁnancing required. With
the standard assumption of βR < 1, liquidity must provide some additional beneﬁt that justiﬁes
the ﬁrm holding liquidity between periods rather than reducing its intertemporal debt, bt, and
associated interest payments. Firms choose to hold liquidity in this economy for two reasons: to
increase their threat point in bargaining over the cost of the intraperiod bank loan and to pay for
labor expenses.

Speciﬁcally, ﬁrms enter the period holding capital for use in production; however, in order
to produce, they must also hire labor at the beginning of the period.
If ﬁrms enter the period
holding less liquidity than necessary to cover desired labor expenses, then they can pay for these
labor expenses by borrowing via an intraperiod bank loan, lt. The ﬁrm and lender bargain over
repayment on the intraperiod loan, i.e. the net cost et per unit of loan. This reveals two beneﬁts to
a ﬁrm from carrying liquidity. First, holding liquidity reduces the size of the intraperiod loan that
a ﬁrm desires, all else equal. Second, as detailed below, holding liquidity increases the value of

23

the ﬁrm’s threat point in the process of bargaining over loan repayment et. The eﬀect of both is to
decrease the total cost of ﬁnancing labor expenses.

To make the solution of the bargaining problem tractable, the cost et is paid by the ﬁrm after
production. Additionally, the ﬁrm has the choice to defer until the end of the period payment on
a fraction 1 − ν of its labor expenses that are paid out of accumulated liquidity.30 Thus, labor
expenses can be written as

wtnt = lt + νat + (1 − ν)at

(1.7)

where wt is the wage rate paid to labor. The ﬁrm and lender bargain over the cost of the intraperiod
loan. Instead of reaching an agreement with the lender, the ﬁrm can threaten to walk away and
produce using only the labor it can aﬀord to hire with its accumulated liquidity. This leads to the
following bargaining problem:

(cid:40)(cid:20)

(cid:18)

(cid:19)

(cid:18)

(cid:19)(cid:21) η(cid:20)

(cid:21)1−η(cid:41)

max
et

F

zt, kt,

lt + at

wt

− (1 + et)lt − F

zt, kt,

at
wt

(et − γ)lt

where η is the bargaining power of the ﬁrm and γ is the return on the lender’s outside option.31
Since the returns of the production function are diminishing in labor, a ﬁrm with higher liquidity,
at, will produce less additional surplus from agreeing to an intraperiod loan. This means, all else
equal, that the cost of the intraperiod loan will be lower for ﬁrms holding more liquidity. Solving
this bargaining problem, the cost of the intraperiod loan is

(cid:20)

(cid:18)(cid:18) lt +at

wt

(cid:19)1−θ −

(cid:18) at

wt

(cid:19)1−θ(cid:19)

(cid:21)

− lt

lt

(1 − η)

et =

zt k θ
t

+ ηγlt

(1.8)

The ﬁrm’s intertemporal budget constraint can be written as follows:

(1 + et)lt + wtnt + bt + kt+1 + ϕ(dt) + at+1 = (1 − δ)kt + F(zt, kt, nt) +

bt+1
Rt

+ at + lt

30See the end of this subsection for further discussion of the ν parameter.
31See the end of this subsection for further discussion of the γ parameter.

24

Then, cancelling lt, which is paid back within the same period as it is contracted, substituting in
and lt = wtnt − at gives the following budget
Equation (1.8) for etlt, substituting in nt = lt +at
wt
constraint:

η(1 + γ)wtnt + bt + kt+1 + ϕ(dt) + at+1

= (1 − δ)kt + ηF(zt, kt, nt) +

bt+1
Rt

+ η(1 + γ)at + (1 − η)F(zt, kt,

)

at
wt

(1.9)

Finally, since ﬁrms are able to default on their loans (i.e.
the enforceability of loan obligations
is imperfect), the ability for a ﬁrm to borrow is limited. Speciﬁcally, at the end of the period,
the ﬁrm can choose to default on the intraperiod loan lt. After production and paying costs etlt,
wtnt −(1− ν)at, bt, kt+1 and ϕ(dt), the ﬁrm is holding liquid resources equal to lt + at+1 +(1− ν)at.
By assumption, the ﬁrm can defer a portion, (1 − ν)at, of its labor costs to the end of the period. If
the ﬁrm defaults, then the lender is able to recover the full value of the ﬁrm’s non-liquid physical
capital, kt+1, with probability ξt and recover nothing with probability 1 − ξt. However, the ﬁrm
is able to hide its liquid resources, lt + at+1 + (1 − ν)at. It follows that the lender’s enforcement
constraint is:32

ξt(kt+1 − bt+1
1 + rt

) ≥ wtnt − νat

(1.10)
Increasing the amount of intertemporal debt, bt+1, or intraperiod debt, lt = wtnt−at, will tighten the
enforcement constraint. Capital, kt+1, serves as collateral and loosens the enforcement constraint.
Note that, all else equal, holding more liquidity loosens the enforcement constraint through reducing
the desired intraperiod loan amount. As in Jermann & Quadrini (2012), ξt, is an aggregate stochastic
innovation where changes are referred to as a “ﬁnancial shock”.

Before solving for the ﬁrm’s optimization problem, two parameters deserve additional discus-
sion. The ν parameter governs the fraction of at that functionally acts as collateral. This parameter
can be rationalized in at least two ways. First, it could be thought of as the lender having an enforce-

32See Appendix for a complete proof of the derivation of the enforcement constraint.

25

ment mechanism that makes the ﬁrm commit to paying a portion of wages in a timely manner. The
portion of the labor costs that are not reliant on the lender can be deferred. This shares similarities
with the block-bargaining assumption of Petrosky-Nadeau & Wasmer (2013) in which the ﬁrm and
banker form a block to negotiate wages with workers. Alternatively, ν can be interpreted as the
portion of liquidity that the lender can verify, i.e. that the ﬁrm cannot escape with in the event of
default. Since the lender can recoup this fraction of liquidity in the event of default, it functionally
acts as collateral.

The γ parameter governs the value of the lender’s outside option in the event that the ﬁrm and
lender do not agree to an intraperiod loan. Thus, it is assumed the lender can still re-invest the
funds, lt, in the event of a negotiation breakdown, but at a lower net beneﬁt. Alternatively, we could
think that the lender has some superior storage technology that returns a non-zero net beneﬁt.

1.6.2 Firm Decisions

Since the two shocks, productivity and ﬁnancial, are aggregate shocks, we can work with a
representative ﬁrm model. Let V(s;k,b,a) be the cum-dividend value of the ﬁrm, where s is the
aggregate states. The representative ﬁrm’s optimization problem then reads:

V(s; k, b, a) = max

, b(cid:48)
d,n,k(cid:48),b(cid:48),a(cid:48){d + Em(cid:48)V(s(cid:48); k(cid:48)

, a(cid:48))}

(1.11)

subject to

ξ(k(cid:48) − b(cid:48)
1 + r

) ≥ wn − νa

and

ϕ(d) = (1 − δ)k + ηzk θn1−θ +

b(cid:48)
R

+ η(1 + γ)a + (1 − η)zk θ( a
w

)1−θ − b − k(cid:48) − a(cid:48) − η(1 + γ)wn.

The ﬁrst constraint is the enforcement constraint (EC) while the second constraint is the budget
constraint (BC). Let λ and µ be the Lagrange multipliers on the budget constraint and enforcement
constraint, respectively, and m(cid:48) be a stochastic discount factor. The FOC for d gives λ =
1
ϕd(d).

26

(cid:18)

a(cid:48):

Em(cid:48) ·

b(cid:48):

Substituting this in for λ, using the envelope conditions for k, b and a and rearranging terms gives
the FOCs:

(cid:18)

ν(cid:124)(cid:123)(cid:122)(cid:125)

(cid:48)
µ

EC loosening

+

1

ϕd(d(cid:48))

(cid:124)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:123)(cid:122)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:32)(cid:125)
η(1 + γ) + (1 − η)(1 − θ)z(cid:48)k(cid:48)θ( a(cid:48)
w(cid:48))−θ · 1
w(cid:48)

Negotiation Beneﬁt

(cid:19)(cid:19)

=

1
ϕd(d)

(1.12)

Accumulating liquidity yields two beneﬁts. First, it loosens the EC in the next period by µ(cid:48)ν: the
multiplier on the next-period EC times the fraction of liquidity that cannot be absconded from the
lender (and thus acts as collateral). Second, accumulating liquidity also loosens the next-period BC
by the next-period BC multiplier times the “negotiation beneﬁt” terms, i.e. accumulating liquidity
lowers the cost of the intraperiod loan. The cost of accumulating liquidity is a reduction in dividend
payments, tightening the BC this period by

1
ϕd(d) = λ.

1

ϕd(d) · 1

R

=

µξ
1 + r

+ Em(cid:48) ·

1

ϕd(d(cid:48))

(1.13)

Intertemporal borrowing loosens the BC this period, but tightens the EC this period as well as the
BC next period.

(cid:18)

(cid:19)

(cid:27)

(cid:26)(cid:18)

k(cid:48):

Em(cid:48) ·

(cid:19)

1

ϕd(d(cid:48))

· (1 − δ + ηθz(cid:48)k(cid:48)θ−1n(cid:48)1−θ + (1 − η)θz(cid:48)k(cid:48)θ−1( a(cid:48)

w(cid:48))1−θ)

+ ξ µ =

1
ϕd(d)

(1.14)

Purchasing capital loosens the BC next period through liquidation, increased production and
lowered cost e (through decreasing returns to scale of the production function) and loosens the EC
this period as collateral. But, it tightens the BC this period.

n:

(1 − θ)zk θn−θ =

ϕd(d)µ + η(1 + γ)

η

· w

(1.15)

Increasing labor increases production by the marginal product of labor. On the other hand, it
tightens the EC through requiring more l and tightens the BC through the wage payment and
increasing the intraperiod loan cost.

27

1.6.3 Household Decisions and General Equilibrium

There is a continuum of identical households, who maximize expected lifetime utility. Households
consume ct and supply labor nt to ﬁrms. Households also act as the ﬁrms’ shareholders and lend
to the ﬁrms by purchasing bonds bt. Thus, households solve the following optimization problem:

max

nt,bt+1,st+1

βtU(ct, nt)

(1.16)

∞

E0

t=0
subject to

wtnt + bt + st(dt + pt) =

bt+1
1 + rt

+ st+1pt + ct + Tt,

where wt is the wage rate, st are the equity shares, dt are the equity payouts received from owning
1+rt(1−τ) − Bt+1
Bt+1
equity shares, pt is the market price of shares, rt is the interest rate, and Tt =
is
1+rt
the lump-sum tax that funds the tax subsidy, τ, of ﬁrms’ debt.33

The household’s FOCs for nt, bt+1, and st+1, respectively, are:

wtUc(ct, nt) + Un(ct, nt) = 0

Uc(ct, nt) − β(1 + rt)EUc(ct+1, nt+1) = 0

Uc(ct, nt)pt − βE(dt+1 + pt+1)Uc(ct+1, nt+1) = 0

(1.17)

(1.18)

(1.19)

The aggregate states s are productivity z, the liquidation technology ξ (capturing the tightness
of the borrowing constraint), the aggregate capital K, the aggregate bonds B and the aggregate
liquidity A. A general equilibrium is deﬁned as follows:

DEFINITION 1: A recursive competitive equilibrium is deﬁned as a set of functions for (i)
households’ policies ch(s), nh(s), and bh(s); (ii) ﬁrms’ policies d(s; k, b), n(s; k, b), k(s; k, b),
b(s; k, b), and a(s; k, b); (iii) ﬁrms’ value V(s, k, b); (iv) aggregate prices w(s), r(s), and m(s, s(cid:48)); (v)

33For simplicity, we assume in the baseline model that the bank’s proﬁts, etlt, are immediately consumed by the bank.
Alternatively, these proﬁts could be distributed to the households as a lump-sum payment. We show in the Appendix
that this does not have a meaningful impact on the results.

28

law of motion for the aggregate states s’=Ψ(s), such that: households’ policies satisfy conditions
1.17-1.18; (ii) ﬁrms’ policies are optimal and V(s, k, b) satisﬁes the Bellman equation 1.11; (iii) the
wage and interest rates clear the labor and bond markets and m(s, s ˆa(cid:48)) = βUc(c ˆa(cid:48), n ˆa(cid:48))/Uc(c, n);
(iv) the law of motion Ψ(s) is consistent with individual decisions and the stochastic processes for
z and ξ.

1.7 Calibration and Quantitative Analysis

The empirical evidence shows a stark contrast in the cyclical ﬁnancing behavior of small ﬁrms
in the post-ﬁnancial consolidation period relative to the pre-consolidation period. The evidence
suggests that ﬁrms with low bargaining power vis-à-vis their lenders and a high value for the
bank’s outside option (i.e. a weaker relationship between the ﬁrm and lender) are responsible for
the post-consolidation behavior. Thus, in this section, we calibrate the model and then vary the
corresponding parameters, η and γ.

We ﬁrst set these parameters to reﬂect a state of the world prior to ﬁnancial consolidation and
widespread interstate banking: high ﬁrms’ bargaining power vis-à-vis their lenders and a low value
for the banks’ outside option. We are going to see that in this “pre ﬁnancial consolidation” world,
the impulse response functions for equity issuance and liquidity accumulation are consistent with
the cyclicality patterns of large ﬁrms and small ﬁrms in the pre-1999 period. However, when η and
γ are varied to reﬂect a state of the world with ﬁnancial consolidation and widespread interstate
banking (i.e. low ﬁrms’ bargaining power and a high value for the banks’ outside option), then the
impulse response functions for equity issuance and liquidity accumulation are consistent with the
smaller ﬁrms in the post-1999 period. The response of debt issuance remains procyclical for all
parameter values, in line with the empirical ﬁndings.

Speciﬁcally, when η and γ are set to reﬂect the post-1999 period with low ﬁrm bargaining power
vis-à-vis their lenders and a high value for the banks’ outside option, then equity issuance and
liquidity accumulation increase in response to a positive shock (i.e. are procyclical). Importantly,
this result depends on the type of shock. For ﬁnancial shocks (i.e. shocks to ξ), equity issuance and

29

liquidity accumulation remain countercyclical. Instead, it is the response to TFP shocks that are
consistent with the empirical cyclicality results. In Section 1.8, we show evidence that TFP shocks
are in fact what drive the main cyclicality ﬁndings.

1.7.1 Calibration

Table B8 displays the calibrated values of our main parameters. The household utility function is
U(c, n) = ln(c) + α · ln(1 − n). The disutility of work parameter α is set such that hours worked, n,
equals 0.3 in steady state. The Cobb-Douglas production function of the ﬁrm has a capital share
parameter θ equal to 0.36. Capital depreciates at the standard rate of δ equal to 0.025. Debt has
a tax advantage over equity of τ equal to 0.35 to match the 35 percent marginal corporate income
tax rate that was in place for most of our sample period. The nonﬁnancial business sector has an
average quarterly debt to GDP ratio of 3.4 (see Jermann & Quadrini (2012)). Thus, we set the
mean value of ¯ξ to target this steady state debt to GDP ratio. Finally, the parameters that govern
the properties of the TFP shock, σz, and the ﬁnancial shock, σξ, are derived from our empirical
estimates of these shocks. Section 1.8 provides details on the construction of these series.

The calibrated values of the discount parameter, β, and the parameter for the value of liquidity
as collateral, ν, require further attention. The product of these two parameters is key to determining
the ﬁrm’s desire to hold liquidity. If the desire to carry liquidity between periods is too high, then
the bargaining between ﬁrms and banks over the intraperiod loan becomes unnecessary as ﬁrms
can cover all wage costs with accumulated liquidity. Thus, we can produce qualitatively similar
results to our baseline by increasing β and decreasing ν. For example, we can increase β from
0.9 to 0.97 and simultaneously decrease ν from 0.25 to 0.1. We detail the ﬁrm bargaining power
parameter, η, and the bank outside option parameter, γ, below.

1.7.2 Simulated Responses

To evaluate the cyclical properties of the ﬁnancing variables, we subject two diﬀerent steady states
to two types of shocks. The ﬁrst steady state is the pre-ﬁnancial consolidation state, which we

30

refer to as the “stronger borrowers” state. We set the ﬁrm bargaining power parameter, η, to 0.99
and the bank outside option parameter, γ, to 0.01. The bargaining power parameter value of
0.99 approximates the case of full ﬁrm bargaining power, as found, e.g., in Jermann & Quadrini
(2012) and Diamond & Rajan (2001). The bank outside option parameter of 0.01 approximates
the case where the bank simply stores the intraperiod loan funds at zero net beneﬁt in the event
of no agreement with the ﬁrm. This is similar to the assumption of Diamond & Rajan (2001), for
example, where a lender’s only outside option is liquidation.

Table B8 shows the steady state values for select variables. Note that debt issuance and liquidity
accumulation are zero in steady state. Thus, the IRFs will show the absolute (i.e. the percentage
point) deviation for each ﬁnancing variable. Consistent with Jermann & Quadrini (2012), the
ﬁnancing variables are scaled by output. Panel (a) of Figure C4 shows the impulse responses
of debt issuance, equity issuance and liquidity accumulation to a one-time positive productivity
(TFP) shock (z) and a one-time positive ﬁnancial shock (ξ) from this pre-ﬁnancial consolidation
steady state. Debt issuance rises upon impact and equity issuance falls for both positive shocks.
This is consistent with the empirical results of debt issuance being procyclical and equity issuance
countercyclical in the pre-1999 period. Liquidity accumulation essentially does not respond to a
shock. Given that liquidity acts as a buﬀer to increased bargaining costs, ﬁrms with such a high
value of η do not need to respond to shocks by adjusting liquidity, as banks are anyway unable to
extract a meaningful amount of surplus in the bargaining process.

In the second steady state, the post-ﬁnancial consolidation steady state, we reduce the ﬁrm
bargaining power parameter from 0.99 to 0.7 and increase the bank outside option parameter to
0.04. Petrosky-Nadeau & Wasmer (2013) estimates the bargaining power parameter for banks in
the US economy as 0.68, but with a range from 0.37 to 0.98. This change acts to illustrate the
eﬀects of ﬁnancial consolidation on smaller ﬁrms, i.e. those in which we see empirical evidence
that bargaining power and the bank’s outside option matter following the ﬁnancial consolidation
of the 1990s. In Panel (b), the “weaker borrowers” state shows the impulse responses for positive
shocks to this new steady state. For the ﬁnancial shock, the magnitude of the impact on equity

31

issuance and debt issuance is smaller; however, the direction of the response remains the same as
in the pre-ﬁnancial consolidation steady state. In contrast, equity issuance now responds positively
to a positive TFP shock, i.e. equity issuance displays a procyclical pattern upon impact. Similarly,
liquidity accumulation now responds positively.

The diﬀerence in the impulse responses in the post-ﬁnancial consolidation steady state, relative
to the pre-ﬁnancial consolidation state, can be interpreted as an enhanced incentive for ﬁrms to
accumulate precautionary liquidity and, hence, to issue equity. Faced with the prospect of a
relevant surplus extraction by their lending banks, ﬁrms have an increased appetite for liquidity
when productivity rises. To ﬁnance this precautionary accumulation of liquidity, they issue more
equity when a TFP shock hits. A positive ﬁnancial shock, by contrast, relaxes the access to external
ﬁnancing. This reduces the need for precautionary liquidity. We now discuss empirical evidence
for the importance of small ﬁrms’ response to TFP shocks in explaining the change in the cyclicality
of equity issuance and liquidity accumulation.

1.8 Financial and TFP Shocks

As shown in the previous section, the change in the cyclicality of equity issuance and liquidity
accumulation occurs only in response to TFP shocks in the model economy. This yields a testable
implication: we can re-estimate the baseline panel regressions, replacing the cyclical component of
real corporate GDP with measures of TFP and ﬁnancial shocks. To be consistent with the model,
we would expect that the procyclicality of equity issuance and liquidity accumulation for small
ﬁrms is due to TFP shocks, rather than ﬁnancial shocks, during the post-1999 period.

To create the baseline measures of TFP and ﬁnancial shocks, we follow the methodology of
Jermann & Quadrini (2012) and extend their series through 2017. First, to create a time series of
productivity shocks, we compute the Solow residuals of the production function:

ˆzt = ˆyt − θ ˆkt − (1 − θ)ˆnt

(1.20)

where the hat represents the log-deviation from the deterministic trend. The output variable, yt, is
real GDP from the National Income and Product Accounts. The capital variable, kt, is from the

32

Flow of Funds Accounts. The labor variable, nt is the total private aggregate weekly hours from
the Current Employment Statistics survey.

Next, we create the ﬁnancial shock series using the (binding) enforcement constraint from

Jermann & Quadrini (2012):34

ξt

(cid:18)

kt+1 − bt+1
1 + rt

(cid:19)

= yt

(1.21)

The ﬁnancial variable ξt is then computed as the residual. The debt variable is from the Flow of
Funds Accounts.

Finally, as in Jermann & Quadrini (2012), we compute the shocks to z and ξ using the following

autoregressive system:

(cid:169)(cid:173)(cid:173)(cid:171)ˆzt+1

ˆξt+1

(cid:170)(cid:174)(cid:174)(cid:172) = A(cid:169)(cid:173)(cid:173)(cid:171)ˆzt

ˆξt

(cid:170)(cid:174)(cid:174)(cid:172) +(cid:169)(cid:173)(cid:173)(cid:171)z,t+1

ξ,t+1

(cid:170)(cid:174)(cid:174)(cid:172)

(1.22)

Figure C7 plots the estimated series of TFP shocks (z,t+1) and ﬁnancial shocks (ξ,t+1), as well
as the cyclical GDP measure. All series have been standardized to have a mean of zero and unit
variance to more easily evaluate the comovement of each measure.

After having computed TFP and ﬁnancial shocks, we replace the cyclical component of real
corporate GDP in the baseline empirical panel speciﬁcation with the one-year lagged value of these
shocks. Since the ﬁrm ﬁnancing data is reported at the annual level, the contemporaneous shock
value contains information for a shock that occurs (at least partially) after the ﬁnancing decision.
Using the lagged shock avoids this issue. The results are displayed in Table B9 for both large
ﬁrms and smaller ﬁrms, split by the pre-1999 and post-1999 periods. In the pre-1999 period, the
results across ﬁrm size are quite similar: a positive ﬁnancial shock (i.e. a loosening of the ﬁnancial
constraint) is associated with an increase in debt issuance, a decrease in equity issuance and a
decrease in liquidity accumulation. As it becomes easier to borrow, both large and small ﬁrms shift
toward issuing debt and away from issuing equity and accumulating liquidity. This aligns with
the earlier cyclicality results and the standard pecking order theory. Interestingly, TFP shocks are
insigniﬁcant for both ﬁrm sizes and all ﬁnancing variables in the pre-1999 period.

34We recognize that this enforcement constraint diﬀers from the one used in our model. To generate ﬁnancial shocks

comparable to the literature, we used this more common enforcement constraint.

33

In the post-1999 period (i.e. following ﬁnancial consolidation), the ﬁnancing behavior of large
ﬁrms remains qualitatively unchanged; however, smaller ﬁrms see a dramatic change. While debt
issuance remains closely related to positive ﬁnancial shocks, the relationship between ﬁnancial
shocks and equity issuance/liquidity accumulation becomes statistically insigniﬁcant. Positive TFP
shocks are now signiﬁcantly associated with an increase in both equity issuance and liquidity
accumulation. Thus, both the type of shock and the sign of the relationship with the relevant shock
have changed for smaller ﬁrms in the post-1999 period. This matches the change observed for
the cyclicality of equity issuance and liquidity accumulation in the latter period. The importance
of TFP shocks is also consistent with the impulse response functions above, in which a change
in cyclicality for equity issuance and liquidity accumulation occurred for TFP shocks and not for
ﬁnancial shocks.

Finally, we can compare the magnitude of the equity issuance response in our model to the
empirical estimates. In the model, equity issuance for weak borrowers increases by 2% over the
ﬁrst four quarters following a positive TFP shock (an increase of 0.35 from the steady state value of
17.3%). Empirically, we estimate in Table B9 that small ﬁrms’ annual equity issuance increases by
2.43 percentage points in response to a positive TFP shock. This is an increase of 9% relative to the
post-1999 average of 26% for small ﬁrms’ equity issuance. Thus, the model explains approximately
two-ninths, or 22%, of the equity issuance response to TFP shocks in the post-1999 period.

1.9 Investment and Employment

To this point, we have documented a signiﬁcant change in the cyclicality of equity issuance
and liquidity accumulation for smaller ﬁrms. We have additionally uncovered both empirical and
theoretical evidence that consolidation among lenders contribute to explaining these changes. In
this section, we investigate the implications for the investment and employment behavior of smaller
publicly-traded ﬁrms over the business cycle.

As seen in equation 1.14, the ﬁrst order condition for capital, there are three main mechanisms
by which a TFP shock impacts a ﬁrm’s demand for capital: a “Surplus Appropriation Channel”,

34

ηθz(cid:48)k(cid:48)θ−1n(cid:48)1−θ; a “Financial Channel”, (1 − η)θz(cid:48)k(cid:48)θ−1( a(cid:48)
w(cid:48))1−θ; and a “Collateral Channel”, ξ µ.
First, increased productivity leads to higher output. The lender will want to extract this surplus
during the bargaining phase; the ﬁrm’s bargaining power, η, determines how much of the surplus
the ﬁrm can keep. The more surplus the ﬁrm can keep, the higher its demand for capital. Second, as
noted, the liquidity holdings of the ﬁrm are used as the threat point in bargaining with the lender, as
they can be used to hire labor. This beneﬁt of liquidity has a complementary eﬀect with the capital
stock. Thus, capital provides a larger beneﬁt through this ﬁnancial channel for ﬁrms with more
accumulated liquidity. Third, capital beneﬁts the ﬁrm as collateral in the enforcement constraint.
Figure C5 shows the IRFs for each of these three components (Surplus Appropriation, Financial,
Collateral) in response to a positive TFP shock.35 Capital increases more for the stronger borrowers
than for the weaker borrowers via the Surplus Appropriation mechanism. This reﬂects the fact
that the higher bargaining power of stronger borrowers limits the lender’s ability to appropriate the
surplus of additional capital. The main diﬀerence between stronger and weaker borrowers is the
Financial Channel response. For stronger borrowers, their bargaining power is so high that they do
not have an incentive to increase their threat point. The opposite is true for weaker borrowers. Thus,
the channel most closely related to ﬁnancial consolidation increases the sensitivity of the weaker
borrowers. That is, the changing cyclicality in corporate ﬁnancing results in higher investment
sensitivity. Interestingly, the Collateral Channel shows minimal diﬀerence between the two types
of borrowers. Next, we investigate empirically whether ﬁnancial consolidation indeed resulted in
higher sensitivity for smaller ﬁrms.

Given the evidence presented above on the cyclicality of ﬁrm ﬁnancing, we would expect
smaller ﬁrms to reduce equity issuance and liquidity accumulation in response to a negative shock,
but only during the post-1999 period. Table B15 shows the change in the ﬁnancing variables of
interest in the face of a negative shock, i.e. the years with negative growth in the cyclical component
of HP-ﬁltered real corporate GDP (1982, 1986, 1989-1993, 2001-2003, 2007-2009, and 2016).36

35Figure C8 shows the same IRFs for a positive ﬁnancial shock.
36In the Online Appendix, we try three alternative deﬁnitions of a negative shock. First, we use an indicator that is
instead set to 1 in years with negative real corporate GDP growth. Second, we use an indicator that is set to 1 in
years that overlap with the NBER dating of a recession. Third, we use a discrete variable that measures the number

35

As expected, small ﬁrms see a large decline in equity issuance and liquidity accumulation during
years with negative growth (relative to positive growth years) and, as seen in Panel B, this holds
true during the post-1999 period only.

Next, in Table B16 we repeat the above exercise with change in investment and log change in
employment replacing the ﬁnancing variables. For both ﬁrm sizes, investment and employment fall
in years with negative economic growth for each subperiod. However, as revealed by the p-values,
it is only the small ﬁrms that see a signiﬁcant increase in responsiveness from the pre-1999 to the
post-1999 period. Alternatively, we can substitute in investment and employment measures for our
ﬁnancing variables in the baseline panel speciﬁcation to estimate the overall cyclicality. As seen
in Table B10, it is again only small ﬁrms that experience a signiﬁcant diﬀerence from pre-1999 to
post-1999. This suggests that the change in the cyclicality of ﬁnancing for smaller ﬁrms may also
have resulted in increased sensitivity of investment and employment.

To further isolate the Financial Channel, we also split small ﬁrms by their liquidity position
leading into the post-1999 period. Speciﬁcally, “low liquidity position” ﬁrms are those small ﬁrms
with a cash-to-asset ratio in 1996-1998 that was at or below the median. “High liquidity position”
are those small ﬁrms with a cash-to-asset ratio in 1996-1998 above the median. In the terminology
of our model, ﬁrms with high liquidity position should have a higher threat point. Small ﬁrms with a
low liquidity position are in a weaker position to counter the eﬀects of ﬁnancial consolidation; thus,
they should be more sensitive in the post-1999 period. Panel B of Table B10 provides evidence that
this was the case. Firms with a low liquidity position prior to 1999 showed a greater increase in the
sensitivity of investment and employment after 1999. This again points to ﬁnancial consolidation
resulting in higher investment sensitivity for those ﬁrms most aﬀected.

1.10 Conclusion

In recent decades, an intense debate has developed on the consequences of ﬁnancial sector
consolidation. This paper contributes by identifying an important eﬀect of ﬁnancial consolidation

of quarters in a year that overlap with the NBER dating of a recession. All three of these deﬁnitions of a negative
shock produce similar results to Table B15.

36

on the corporate sector, in the form of a structural change in ﬁrms’ ﬁnancing behavior over the
business cycle. We ﬁnd that due to the weakened bargaining power vis-à-vis their lending banks
and a fraying of the relationships between ﬁrms and banks, small and medium-sized publicly-
traded ﬁrms began to issue equity and accumulate precautionary liquidity during expansions. This
behavior starkly contrasts with the countercyclical equity and liquidity behavior of larger publicly-
traded ﬁrms and reﬂects the attempt of small and medium-sized ﬁrms to oﬀset their weakened
position vis-à-vis larger and more complex ﬁnancial institutions. The change in cyclical ﬁnancing
behavior appears to also have far-reaching consequences for ﬁrms’ investment and labor hiring
decisions. The empirical evidence presented in the paper shows that small and medium-sized
ﬁrms’ investment and employment became signiﬁcantly more sensitive to negative shocks.

The paper leaves relevant questions open for future research. For example, equity issuance and
hoarding of precautionary liquidity can entail relevant costs for ﬁrms. Thus, it becomes important
to evaluate the welfare implications of the altered ﬁnancing patterns. Further, as noted, private ﬁrms
are likely to be even more exposed than small publicly listed ﬁrms to ﬁnancial consolidation, as they
lack access to stock markets as a form of ﬁnancing alternative to bank lending. In this sense, the
results of this analysis may constitute a lower bound of the actual eﬀects of ﬁnancial consolidation
through cyclical ﬁnancing patterns. We leave these and other relevant issues to further research.

37

CHAPTER 2

MONETARY POLICY AND FIRM HETEROGENEITY: THE ROLE OF LEVERAGE

SINCE THE FINANCIAL CRISIS

2.1 Introduction

Since the federal funds rate hit the zero lower bound at the beginning of the ﬁnancial crisis,
the Federal Reserve has relied more on unconventional policy tools like forward guidance and
quantitative easing. In this paper we explore how the monetary transmission mechanism may have
changed since the crisis, with a focus on the role of heterogeneity in ﬁrms’ ﬁnancing conditions.
While the importance of the balance sheet of ﬁrms for the monetary transmission mechanism has
long been established, recent work has highlighted the role of ﬁrm-level heterogeneity.1 However,
this literature on ﬁrm-level ﬁnancial heterogeneity has typically focused on the pre-crisis period to
study the transmission of conventional monetary policy actions.2 Our main contribution is to show
that the role of ﬁnancing conditions in determining the ﬁrm-level response to monetary shocks has
reversed in the post-crisis sample.

We document this result with three complementary empirical approaches using both high fre-
quency ﬁnancial market and quarterly real activity data for non-ﬁnancial ﬁrms. For all approaches,
we construct monetary policy shocks using high frequency data from futures and Treasury bond
markets. Our preferred measure of monetary policy shocks combines unexpected changes in the
federal funds target with the change in the 10 year Treasury yield in a narrow window around FOMC
announcements. This allows us to parsimoniously capture both conventional and unconventional
monetary policy actions.3

*This chapter is joint work with Aeimit Lakdawala.
1For an early survey of the importance of the credit channel of monetary policy, see Bernanke & Gertler (1995). For
recent work on ﬁrm-level heterogeneity see Ottonello & Winberry (2018), Jeenas (2018) and Ozdagli (2018).
2A notable exception is the paper of Wu (2018), which we discuss below.
3This approach is especially important in the post-crisis sample where the federal funds rate is mostly stuck at the zero
lower bound. But we also show that using one single policy indicator for both the pre- and post-crisis samples (e.g.
the change in the 2 year Treasury rate) conﬁrms our results.

38

For our ﬁrst approach, we combine ﬁrm-level characteristics with high frequency data on stock
prices. Using leverage as the measure of the ﬁrm’s ﬁnancial position we ﬁnd that before the ﬁnancial
crisis of 2007-09, stock prices of ﬁrms with higher leverage respond less to monetary policy shocks
on FOMC announcement days. However, this pattern is reversed after the crisis: in the post-crisis
sample ﬁrms with higher leverage respond more to monetary shocks. The panel data allows us to
control for a variety of ﬁrm level variables including a ﬁrm ﬁxed eﬀect to account for any permanent
features at the ﬁrm level. More importantly, since we are interested in the interaction of leverage
and monetary shocks, with time ﬁxed eﬀects we can also control for any aggregate factors that
could be changing over time.

Given that monetary policy shocks are not predictable, these results have no implications for
the expected direction of the movement in the stock price of ﬁrms with higher (or lower) leverage
on FOMC announcement days. However, there is a direct implication for the expected volatility of
the stock price. Speciﬁcally, we should expect that ﬁrms with high leverage will be less volatile on
FOMC announcement days in the pre-crisis sample. Moreover, this relationship should ﬂip with
the crisis making high leverage ﬁrms more volatile on announcement days in the post-crisis sample.
Our second approach involves testing this hypothesis by using high frequency ﬁrm-level options
data. These options data allow us to construct a measure of expected volatility for each ﬁrm. We
analyze these ﬁrm-level expected volatility measures on the day before the FOMC announcement
and conﬁrm the reversal in the relationship between leverage and monetary policy announcements
since the ﬁnancial crisis. Markets expected high leverage ﬁrms to be less volatile on FOMC
announcement days in the pre-crisis sample but more volatile since then.

Our third approach involves using ﬁrm-level investment data. Since this measure of real
activity is only available quarterly, we aggregate our monetary policy shock measure up to the
quarterly level. At this quarterly frequency, there are several factors that could aﬀect ﬁrm-level
investment other than monetary policy.4 Nevertheless, these quarterly results conﬁrm the pattern

4Since we use an exogenous measure of monetary shocks and time-ﬁxed eﬀects we are not worried about endogeneity
but rather about the loss of precision as a smaller fraction of the ﬁrm-level dependent variable will be driven by
monetary policy here relative to the high frequency speciﬁcation.

39

of increasing responsiveness of ﬁrms with higher leverage since the ﬁnancial crisis. There is an
ongoing discussion in the literature regarding the longer-run response of investment to monetary
policy shocks, see Ottonello & Winberry (2018) and Jeenas (2018). The relatively shorter sample
of data since the crisis makes it diﬃcult to do inference on comparing long-run responses in the pre-
and post-crisis samples, so in this paper we focus on the contemporaneous response. Our ﬁndings
for the contemporaneous response of investment in the pre-crisis sample are consistent with both
these papers.

Our results hold across a variety of robustness checks, including using alternative measures of
leverage, expanding our baseline sample from ﬁrms in the S&P 500 to a broader set of ﬁrms in
the CRSP/Compustat dataset, dropping unscheduled FOMC meetings, using time-by-sector ﬁxed
eﬀects and including ﬁnancial ﬁrms in our sample. A natural question is whether our results are
driven by the changing behavior of leverage since the crisis. We document that average leverage
has only slightly increased since the crisis and that the cross-sectional distribution of leverage is
similar in the two samples. Moreover, we show that most ﬁrms have not moved around much in the
leverage distribution since the crisis and that excluding ﬁrms that did move around a lot does not
aﬀect our results.

In the next part of the paper we shed light on the mechanism driving our empirical results.
There is a growing literature on the transmission of monetary policy and heterogeneity in ﬁrm
balance sheets. But, as mentioned above, this literature has focused on the pre-crisis period. We
start by placing our pre-crisis results in the context of the leading heterogenous ﬁrm model of
Ottonello & Winberry (2018), which builds on the work of Khan et al. (2016). Within this model,
we then discuss potential channels that can rationalize our post-crisis results and provide supporting
empirical evidence.

In the Ottonello & Winberry (2018) model there are competing forces aﬀecting how high vs.
low leverage ﬁrms respond to an expansionary monetary policy shock. The marginal cost curve of
a high leverage ﬁrm is steeper making it less responsive to monetary policy induced shifts of the
marginal beneﬁt curve. However, the expansionary monetary policy shock ﬂattens out the marginal

40

cost curve more because of an increase in the value of collateral and cash ﬂows, making high
leverage ﬁrms more responsive. In a pre-crisis calibration, Ottonello & Winberry (2018) ﬁnd that
the former eﬀect dominates, implying results that are consistent with our pre-crisis ﬁndings. We
argue that to reverse this result in the post-crisis sample, an expansionary monetary policy shock
must ﬂatten the marginal cost curve substantially more for high leverage ﬁrms. A direct testable
implication of this hypothesis is that credit spreads for high leverage ﬁrms should fall more (relative
to low leverage ﬁrms) in the post-crisis sample. We provide evidence of this channel by using the
credit spread between ﬁrms rated BAA relative to those rated AAA. Consistent with the hypothesis,
the BAA-AAA spread falls more (in response to an expansionary monetary policy shock) in the
post-crisis sample.5

Why is it that monetary policy actions ﬂatten the marginal cost curve more and thus compress
the yield spread between high leverage and low leverage ﬁrms more in the post-crisis sample? Our
hypothesis is that longer-term interest rates, and thus ﬁrms that are more dependent on long-term
funding, have become more sensitive to monetary policy actions in the post-crisis sample. We ﬁrst
document that the nominal and real 10 year Treasury yields respond signiﬁcantly more to monetary
policy shocks in the post-crisis sample. Next, we test if this increased sensitivity is spilling over
into long-term funding conditions and contributing to our baseline results. To do this we separate a
ﬁrm’s leverage into two components, one depending on long-term debt and the other on short-term
debt. We ﬁnd that the increased responsiveness since the crisis is largely driven by ﬁrms that have
a larger share of long-term debt in their leverage. This is consistent with the results of Foley-Fisher
et al. (2016) who ﬁnd that ﬁrms that are more dependent on long-term debt responded more to
the Federal Reserve’s Maturity Extension Program implemented in 2011 and 2012. However,
our results suggest that this increased responsiveness to monetary policy is a feature prevalent
throughout the post-crisis sample rather than just in response to speciﬁc large scale asset purchase
episodes.

We also provide some suggestive evidence for the role of another monetary transmission channel,

5This result is broadly consistent with the ﬁndings of Gilchrist & Zakrajšek (2013) that borrowing costs of riskier ﬁrms
reacted substantially more to quantitative easing announcements in the post-crisis sample.

41

namely through changes in uncertainty about future policy decisions. In recent empirical work,
Kroencke et al. (2018), Bauer et al. (2019) and Bundick et al. (2017b) ﬁnd that monetary policy
uncertainty shocks are likely drivers of term-premium on long rates and general risk-premium in
the ﬁnancial markets. Moreover, Bauer et al. (2019) ﬁnd that monetary transmission to ﬁnancial
markets through uncertainty has strengthened in the post-crisis sample. We use their uncertainty
measure of the changes in the standard deviation of the expected future policy rate, as constructed
using high-frequency options data around FOMC announcements. We ﬁnd that monetary policy
uncertainty shocks induce a similar reversal in the sign of the relationship between the ﬁrm-level
response and leverage. Speciﬁcally, in response to uncertainty shocks ﬁrms with high-leverage
respond less in the pre-crisis sample but more in the post-crisis sample.

Related Literature: Our paper is related to three strands of the literature. The ﬁrst one
identiﬁes ﬁrm-level characteristics, particularly ﬁnancial constraints such as leverage, that are
associated with a heterogenous stock market response to monetary policy shocks. Both Ehrmann
& Fratzscher (2004) and Ottonello & Winberry (2018) ﬁnd that ﬁnancial constraints aﬀect the
strength of a ﬁrm’s response to monetary policy. Consistent with our results, they ﬁnd evidence
that stock prices for ﬁrms with high leverage are relatively less responsive to monetary shocks
in the pre-crisis period. Ozdagli (2018) ﬁnds that ﬁrms that have higher information frictions
are less responsive while Ippolito et al. (2018) ﬁnd that more ﬁnancially constrained ﬁrms have
a stronger response to monetary policy.6 While most of the literature focuses on the period
prior to the ﬁnancial crisis, Wu (2018) analyses stock price responsiveness to monetary policy
during the 2008-2012 period. Consistent with our results, he ﬁnds that ﬁrms with higher leverage
were more responsive to monetary policy during this period. Our contribution is to highlight the
changing relationship between leverage and stock price response since the ﬁnancial crisis. We also
conﬁrm this changing relationship using both high-frequency options data and quarterly ﬁrm-level
investment data. Finally, we interpret our ﬁndings through a structural model and provide evidence
of the mechanism working through long-term funding conditions and monetary policy uncertainty

6Ippolito et al. (2018) ﬁnd increased sensitivity using a measure of leverage that only includes bank debt.

42

changes.

Our paper also adds to the growing literature on the heterogenous eﬀects of unconventional
monetary policy since the crisis. In addition to the Foley-Fisher et al. (2016) paper discussed above,
there is some recent work analyzing the heterogenous impact of European Central Bank’s (ECB)
unconventional policies. Grosse-Rueschkamp et al. (2019) study the eﬀect of ECB’s corporate
sector purchase program on ﬁrm’s capital structure, while Daetz et al. (2018) investigate the impact
of the ECB’s longer-term reﬁnancing operations on corporate investment.

Finally, our paper is related to the literature that explores heterogenous responses of real
economic activity to changes in monetary policy. Gertler & Gilchrist (1994), an early inﬂuential
paper in this literature, notes that sales at small manufacturing ﬁrms decrease disproportionately
relative to larger manufacturing ﬁrms after Romer and Romer tight money dates.7 They provide
evidence that small ﬁrms are a proxy for ﬁnancial constraints, as smaller ﬁrms seem to have more
diﬃculty acquiring credit when monetary policy becomes contractionary. More recent papers
explicitly attempt to control for ﬁnancial constraints. Ottonello & Winberry (2018) ﬁnd that
investment spending at ﬁrms with higher leverage is less responsive to monetary policy shocks in
the quarter of a monetary shock. Dedola & Lippi (2005) show that output of industries in the U.S.
and four other OECD countries with higher leverage is less responsive between 4 and 12 quarters
after a monetary policy shock. In contrast to those two papers, Jeenas (2019) and Jeenas (2018)
ﬁnd that sales and investment of higher leverage ﬁrms are more responsive to monetary policy
shocks after approximately 8 quarters. We provide evidence that the contemporaneous eﬀect on
higher leverage ﬁrms has become larger following the ﬁnancial crisis. Cloyne et al. (2018) stress the
importance of ﬁrm age for the investment response to monetary shocks with younger ﬁrms being
the most responsive. In our analysis the combination of ﬁrm- and time-ﬁxed eﬀects eﬀectively
control for age and imply that ﬁrm age is not driving the changing relationship between leverage
and monetary transmission.

The rest of the paper is organized as follows. In the next section we outline the data sources

7A related literature uses business cycle contractions as the type of shock under investigation rather than monetary
shocks, e.g. Kudlyak & Sanchez (2016), Crouzet et al. (2017), Kalemli-Ozcan et al. (2018) and Bustamante (2018).

43

and data variables used in our empirical analysis. Section 2.3 presents the results from our three
main empirical strategies using stock price, options and investment data. Next, in Section 2.4 we
shed some light on the mechanism driving our results. Section 2.5 provides a variety of robustness
checks and Section 2.6 concludes.

2.2 Data

This paper uses the daily ﬁrm share prices from the CRSP/Compustat Merged Security Daily
dataset for July 1991 to December 2017 and ﬁrm characteristics from the 1991:Q3 to 2017:Q4
CRSP/Compustat Merged Fundamentals Quarterly dataset. We combine this ﬁrm-level data with
measures of monetary policy shocks that occur on FOMC meeting days. Additionally, we merge a
subsample of the ﬁrms in the CRSP/Compustat Merged dataset with a dataset of ﬁrm-level implied
volatility from OptionMetrics. This section further describes these three data sources.

2.2.1 Monetary Policy Shocks

To construct our measure of monetary policy shocks, we combine data from fed funds futures
and Treasury bond markets. In the high-frequency monetary policy literature, the most common
method to construct shocks involves looking at the change in futures contracts around FOMC
announcements, where the underlying asset is the fed funds rate. The early work of Kuttner (2001)
and Bernanke & Kuttner (2005) used changes in the current month’s futures contract. This measure
captures any unexpected changes to the target for the fed funds rate. However, in more recent years,
the Federal Reserve has been using alternative unconventional policy tools, including large scale
asset purchases (quantitative easing) and forward guidance. FOMC announcements that provide
information about these unconventional policy actions are not well captured by this measure. This
issue has been especially relevant since the fed funds rate hit the zero lower bound in late 2008.
Thus we also use the change in longer-term Treasury yields around FOMC announcements to
supplement the Kuttner (2001) measure. Speciﬁcally, our monetary policy shock m
is deﬁned as
t

t = Pt+δ+ − Pt−δ−
m

44

(2.1)

where t is the time of the FOMC announcement, Pt is either the implied fed funds rate from the
price of the current month’s fed funds futures contract or on-the-run Treasury yields, t + δ+ and
t − δ− represent 20 minutes after the FOMC announcement and 10 minutes before the FOMC
announcement, respectively. For our baseline measure (labeled MP Shock) we combine the change
in the current month’s futures contract8 (labeled FFR Shock) and the 10 year Treasury yield by
taking the ﬁrst principal component of these two measures. The idea is to parsimoniously capture
both conventional and unconventional monetary policy actions in one tool.9 Since the scale of
this shock is arbitrary, we rescale it to have a unit eﬀect on the 2 year yield. We also present our
baseline regressions including both the FFR Shock and the change in the 10 year yield as separate
measures of monetary policy shocks.10 Finally, we also use a single monetary policy indicator in
both the samples: the change in the 2 year Treasury yield as recommended by Hanson & Stein
(2015a). We ﬁnd that the main results are robust to the choice of monetary policy shock; however,
for simplicity, we most frequently report results based on the MP Shock, i.e.
the ﬁrst principal
component of changes in the current month’s fed funds futures contract and the on-the-run 10-year
Treasury yield.

Table D1 shows the summary statistics for the monetary policy shock measures for a pre-crisis
sample (July 1991 to June 2008) and a post-crisis sample (August 2009 to December 2017). The
eﬀect of the zero lower bound is clearly apparent. The standard deviation of the FFR Shock measure
falls substantially from 9 basis points in the pre-crisis sample to 1 basis point in the post-crisis
sample. Even for the two year shock measure the standard deviation falls from 6.5 basis points
to 3.5 basis points, reﬂecting the eﬀective lower bound on Treasury yields starting in late 2011 as
reported by Swanson & Williams (2014). However the standard deviation of the 10 year shock

m(t)
τn
m(t)−τd
τn

m(t) where τn

8Since fed funds futures contracts are based on the average rate for a month, the change in the implied rate needs to be
m(t) is
adjusted by τ(t) =
the day of the month the announcement occurred
9We perform this principal component analysis separately for the pre-crisis and post-crisis periods. The ﬁrst principal
component explains 84% of the variation in these two rates during the pre-crisis sample and 90% of the variation in
these two rates during the post-crisis sample.
10Since the 10-year rate is a function of shorter-term rates, we ﬁrst regress it on the short-term rate and use the residuals

m(t) is the number of days in the month of the announcement and τd

as the orthogonalized measure of the 10-year rate.

45

measure is roughly similar in the pre- and post-crisis samples, motivating our reliance on this
measure to eﬀectively capture monetary policy shocks in the post-crisis period.

2.2.2 Firm-Level Variables

We use the CRSP/Compustat Merged Fundamentals Quarterly sample beginning in 1991:Q3, as
this is the ﬁrst year where we have a complete record of our monetary policy shock measures. For
the baseline results we use the ﬁrms in the S&P 500 index and, as is common in the literature, we
exclude ﬁnancial ﬁrms (SIC 6000-6999). In the Online Appendix we include robustness checks
expanding the sample to all ﬁrms in the CRSP/Compustat dataset and also another check including
ﬁnancial ﬁrms in our sample. Our primary measure of interest from Compustat is the ﬁrm’s leverage
ratio. The baseline results use the ratio of debt-to-capital, measured as the sum of debt in current
liabilities (Compustat item: DLCQ) and long-term debt (DLTTQ) over the sum of debt in current
liabilities, long-term debt and stockholder’s equity (SEQQ). We also conﬁrm our results (relegated
to the Online Appendix) using an alternative measure of leverage: debt-to-assets (using the book
value of assets (ATQ)). Table D1 displays the summary statistics for these deﬁnitions of leverage
measured as the 4-quarter rolling average at the ﬁrm level.

Additionally, we create several control variables using these quarterly data: year-over-year real
sales growth, ﬁrm size as measured by the log of the book value of assets, price-to-cost margin,
receivables-minus-payables to sales, depreciation to assets, ﬁrm age, the log of quarterly market
capitalization and the ratio of current assets to total assets. Including these controls are intended
to capture important characteristics of the ﬁrm that could be correlated with both ﬁrm leverage
and ﬁrm performance. Table D10 in the appendix displays summary statistics of these measures,
dividing the sample into ﬁrms with above and below average leverage. The construction of these
variables follows standard methods in the literature; however, we include these details in the Online
Appendix.11

11In Section 2.3.4 we investigate the impact of monetary policy shocks on ﬁrms’ quarterly investment. The construction

of this investment variable is also detailed in the Online Appendix.

46

We also use daily stock returns and implied volatility measures at the ﬁrm level. We use the
daily return of a ﬁrm’s share price on the day of an FOMC meeting, measured as the log diﬀerence
between the closing share price on the day of the FOMC meeting and the closing share price on the
day prior to the FOMC meeting. The implied volatility measures are computed using ﬁrm-level
options data from OptionMetrics. The methodology used to do this calculation closely follows the
one used for implied volatility of the S&P 500 index, i.e.
the VIX. This daily data is available
from January 1996 to December 2017. To ensure suﬃcient liquidity in the market for options, we
make two restrictions to our implied volatility sample. First, we use implied volatility measures
for options set to expire in greater than 3 months. Second, we use the 100 most-liquid ﬁrms, i.e.
the 100 ﬁrms with the highest number of days with available data in our sample period. Summary
statistics for both these variables are also presented in Table D1.

2.3 Results

This section presents the main results illustrating how leverage aﬀects the ﬁrm-level response to
monetary shocks and how that relationship has changed since the ﬁnancial crisis. First we document
this changing eﬀect using high frequency data on stock prices. Next, we use ﬁrm-level options
data to show that ﬁnancial market participants have been aware of this changing responsiveness.
Finally, we use quarterly data on ﬁrm investment and show that a similar pattern emerges.

2.3.1 Evidence from ﬁrm-level stock returns

We ﬁrst examine how leverage determines the stock price response to monetary policy shocks. In
our baseline results we will consider a pre-crisis sample ranging from July 1991 to June 2008 and a
post-crisis sample from August 2009 to November 2017. We are thus leaving out the crisis period
as categorized by July 2008 to July 2009. These dates are commonly used in the literature due
to turbulence in the ﬁnancial markets and the presence of some asset pricing anomalies, see for
example Nakamura & Steinsson (2018).12

12In the Online Appendix we show that our results are robust to including the ﬁnancial crisis dates in the “post-crisis"

sample.

47

Our baseline regression takes the following general form

si,t = αi + αt + βli,t−1m

t + δli,t−1 + Γ(cid:48)Zi,t−1 + ei,t

(2.2)

where si,t is the (daily) return on ﬁrm i’s share price on FOMC meeting day t,13 αi is a ﬁrm i ﬁxed
eﬀect, αt is an FOMC meeting day t ﬁxed eﬀect (i.e. a dummy for each time period), li,t−1 is ﬁrm
i’s average leverage (measured as debt-to-capital) for the four quarters preceding the quarter of the
FOMC announcement, m
is the monetary policy shock, and Zi,t−1 is a vector of ﬁrm-level controls
t
(lagged by a quarter). The monetary policy shock m
is not included separately as a regressor
t
because it is subsumed by the time ﬁxed eﬀect. Zi,t−1 includes the following ﬁrm-level ﬁnancial
measures as controls: real sales growth, the log of the book value of assets, the price-to-cost margin,
receivables-minus-payables to sales, depreciation to assets, ﬁrm age, the log of quarterly market
capitalization and the ratio of current assets to total assets.14 Since the ﬁrm-level characteristics are
measured at the quarterly level, the leverage ratio and the ﬁrm-level controls are lagged to ensure
they are predetermined at the time of the FOMC announcement. We also include a dummy for the
ﬁscal quarter, to account for diﬀerences across ﬁrms due to diﬀerent positions in their ﬁscal year.
The ﬁrm ﬁxed eﬀect accounts for permanent characteristics of the change in ﬁrm i’s stock price that
are not captured by our controls. The time ﬁxed eﬀect accounts for aggregate shocks common to all
the ﬁrms on the day of the FOMC announcement. The standard errors reported in the parentheses
are calculated using two-way clustering along the time and ﬁrm dimensions.15

We multiply the monetary policy shock measure by negative one so that an increase in m
t
corresponds to an expansionary shock. The key parameter in the above speciﬁcation is β, which
captures how the responsiveness of a ﬁrm’s share price to a monetary policy shock changes based
on a ﬁrm’s leverage ratio. We standardize leverage to be mean zero and unit variance, so β can be
interpreted as the additional change in a ﬁrm’s daily stock price in response to a unit expansionary
monetary shock by moving from an average level of leverage to one standard deviation above the
13si,t = ln(pi,t) − ln(pi,t−1) where the stock price p is measured at the end of the day.
14In the Table D11 we show that our results are robust to interacting these controls, as well as the ﬁrm’s sector, with
the monetary policy shock.

15Our results are robust to using Driscoll-Kraay standard errors instead of two-way clustering.

48

average leverage. In standardizing leverage we use the full sample mean and standard deviation of
leverage across all ﬁrms.

Table D2 presents the results for ﬁrms in the S&P 500 for pre- and post-crisis samples for
the three diﬀerent monetary policy shock measures. Columns 1a and 1b show the results for
our ﬁrst measure (labeled MP shock), which is the ﬁrst principal component of the change in the
current month’s fed funds futures contract and the change in the on-the-run 10 year yield. The
interaction coeﬃcient of MP shock and leverage (β) is negative and signiﬁcant in the pre-crisis
sample but positive and signiﬁcant in the post-crisis sample. Since we use time ﬁxed eﬀects, we
cannot estimate the stand-alone eﬀect of the monetary policy shock on ﬁrm-level stock returns
from this speciﬁcation. However, we show in Table D12 that the sign is as expected and implies
that a 100 basis point expansionary monetary policy shock leads to an over 6% increase in stock
prices.16 This means that the interaction coeﬃcient of MP shock and leverage shows that ﬁrms
with higher leverage were less responsive to monetary shocks before the crisis and have become
more responsive since the crisis. Speciﬁcally, for a ﬁrm that has leverage one standard deviation
above the mean, its stock price falls by 5.5% more (relative to a ﬁrm that has average leverage) in
the pre-crisis sample but increases by 2.2% more in the post-crisis sample. To formally test that
the pre- and post-crisis responses are statistically signiﬁcantly diﬀerent from each other we run a
speciﬁcation with triple interactions (Dpost
is a dummy for the post-crisis
sample. The lower panel of Table D2 shows that the coeﬃcient on this triple-interaction is positive
and statistically signiﬁcant implying that ﬁrms with higher leverage have become signiﬁcantly more
responsive to monetary shocks in the post-crisis, relative to the pre-crisis.

∗ m
t ∗ li,t−1) where Dpost

t

t

Columns 2a and 2b show the results where we include both the change in the current month’s
fed funds futures contract (FFR shock) and the change in the on-the-run 10 year yield as monetary
shocks.17 In the pre-crisis sample ﬁrms with higher leverage are less responsive to a FFR shock
while the 10 year shock response is essentially the same for ﬁrms regardless of leverage. In the

16Recall that the MP shock measure has been scaled to have a unit (100 basis point) eﬀect on the 2 year Treasury yield.
17Since the 10 year yield is mechanically a function of the short rate, we ﬁrst orthogonalize the 10 year yield change
with respect to the FFR shock. In other words, we ﬁrst regress the change in the on-the-run 10 year yield on the FFR
shock and use the residual from this regression as our monetary shock measure (labeled “10 yr shock").

49

post-crisis sample this relationship changes and the FFR shock has an identical eﬀect for ﬁrms
across the leverage spectrum whereas ﬁrms with higher leverage are now responding more to the
10 year shock. The heterogenous eﬀects of monetary policy worked directly through fed funds rate
target changes in the pre-crisis sample but in the post-crisis sample worked through the changes
in the 10 year rate. This is not surprising given that the fed funds rate was stuck at the zero lower
bound for most of the post-crisis sample and in this period the Federal Reserve used unconventional
measures like forward guidance and quantitative easing. This result also highlights the convenience
of our MP shock measure that parsimoniously captures the joint eﬀect of the FFR shock and the 10
year shock in one variable. Thus we will use the MP shock for most of the results shown below.

Finally, Table D2 shows the results when we use only a single rate as the monetary policy
indicator. Columns 3a and 3b show the results using the change in the on-the-run 2 year Treasury
yield as the monetary policy shock (labeled “2 yr shock") for both samples. These results conﬁrm
the reversal in the relationship. Firms with higher leverage were less responsive to monetary policy
shocks in the pre-crisis sample but more responsive in the post-crisis sample. The bottom panel
also shows the triple interaction speciﬁcation using the 2 year rate as the monetary policy indicator.
Consistent with the earlier results, this interaction coeﬃcient is positive and statistically signiﬁcant.
In Section 2.5 below we conduct a battery of robustness checks. These include using a diﬀerent
leverage measure (debt-to-assets), expanding our S&P 500 sample to all non-ﬁnancial ﬁrms in the
CRSP/Compustat dataset, including the crisis dates in the sample, excluding unscheduled FOMC
meetings from the sample, using time-by-sector ﬁxed eﬀects, narrowing our panel to ﬁrms without
any entry or exit from the sample and including ﬁnancial ﬁrms. First, we document some empirical
patterns in our baseline leverage measure to show that our results are not being driven by any
sudden change in the behavior of leverage since the crisis.

2.3.2 Leverage in the pre- and post-crisis samples

The results of diﬀerential responsiveness in the pre- versus post-crisis samples raise some natural
questions about the behavior of leverage in the two samples. Has average leverage changed since

50

the crisis? How does the cross-sectional distribution of leverage compare across the two samples?
Has there been any “churning” of ﬁrms from low leverage in one sample to high-leverage in the
other sample? Importantly, do these patterns play a role in driving the results? In this section we
tackle these issues in order.

First, from Table D1 we can see that leverage is on average only slightly higher in the post-crisis
sample. For example our baseline measure of leverage, debt-to-capital, has a mean of 0.41 in the
post-crisis sample relative to a mean of 0.38 in the pre-crisis sample.18 Similarly the standard
deviation of leverage is also roughly the same across the two samples. Figure E1 shows the
distribution of leverage in the two samples where we have taken the ﬁrm-speciﬁc average for each
sample. The grey shaded bars show the histogram for the pre-crisis sample while the red transparent
bars show the post-crisis histogram. While there is a little more mass toward the right in the post-
crisis sample (and a little more toward the left in the pre-crisis sample), the distribution is quite
similar in the two samples. In our baseline results presented in Section 2.3.1 we standardized our
leverage measure by using the full sample mean and standard deviation of leverage. We have also
tried using the pre-crisis mean and standard deviation to standardize our leverage measure. These
results are presented in the Online Appendix. As one would expect with the patterns from Table
D1 and Figure E1 we ﬁnd these results are very similar to our baseline results.

We further investigate whether ﬁrms have moved around in the distribution in the two samples.
Given the stability of the leverage distribution in the two samples, it is still possible that our results
are driven by i) less-sensitive ﬁrms that had high-leverage in the pre-crisis sample but switched
to having lower leverage in the post-crisis sample and ii) more-sensitive ﬁrms with low leverage
in the pre-crisis sample but switched to having higher leverage in the post-crisis sample. To this
end, Figure E1 displays a scatter plot of the ﬁrm-speciﬁc average leverage in the post-crisis sample
versus the average in the pre-crisis sample. If ﬁrms’ leverage across the two samples is similar, we
should expect the points in the scatter plot to cluster around the 45 degree line. Figure E1 does
in fact show this pattern. We also investigate whether our results are driven by the ﬁrms that did

18The fact that leverage is increasing a little on average for all ﬁrms in the post-crisis sample is not a concern because

our results are driven by how ﬁrm response is diﬀerent across the leverage distribution.

51

change their leverage noticeably, i.e. the ones that are not close to the 45 degree line. In the Table
D13, we present our baseline results excluding ﬁrms which lie more than 1 standard deviation away
from the 45 degree line. The table conﬁrms that our baseline stock market results are robust to
excluding these outliers. This suggests that movement of ﬁrms across the leverage distribution does
not explain the diﬀerence in transmission of monetary policy through ﬁrm leverage following the
ﬁnancial crisis.

Next, we show this pattern of high leverage being related to less responsiveness before the

ﬁnancial crisis and more responsiveness afterwards is also evident using ﬁrm-level options data.

2.3.3 Evidence from ﬁrm-level options data

Given that monetary policy shocks are not predictable, our results from Section 2.3.1 have no
implications for the expected direction of the movement in the stock price of ﬁrms with higher
(or lower) leverage on FOMC announcement days. However, there is a direct implication for the
expected volatility of the stock price of ﬁrms with higher (or lower) leverage. Speciﬁcally, we
should expect that in the pre-crisis period high leverage ﬁrms should be less volatile on FOMC
announcement days but more volatile in the post-crisis sample. In this section, using options data
we indeed ﬁnd strong evidence for this pattern.

We construct ﬁrm-level measures of expected volatility using options data from the Option-
Metrics dataset. The methodology used to do this calculation closely follows the one used for
implied volatility of the S&P 500 index, i.e. the VIX. Speciﬁcally, for each ﬁrm we use the implied
volatility of the expected stock return, based on ﬁrm-level options prices. To assure the options in
our sample are suﬃciently liquid, we restrict the sample to those options set to expire in one quarter
or longer. Even with this restriction, on any given trading day there exist many ﬁrms with missing
implied volatility data. Thus, for the following speciﬁcation, we will use the the most recent day
(within the previous three trading days) in which an S&P 500 ﬁrm has a non-missing value for its
implied volatility.19

19Our results are very similar if we instead choose not to impute any of the missing implied volatility data

52

Using this implied volatility measure we explore whether the interrelation between ﬁrm-level
expected volatility, leverage and FOMC announcements has changed in a way that is consistent
with our earlier results. Speciﬁcally we run the following regression

ivoli,t−1 = αi + αt + δli,t + βli,t−1Dpost

t

+ ΓZi,t−1 + ei,t

(2.3)

where for an FOMC meeting occurring on day t, ivoli,t−1 is the level of implied volatility for ﬁrm
i on the day before the FOMC meeting, li,t−1 is average leverage (debt-to-capital) for ﬁrm i for the
four quarters preceding the quarter of the FOMC announcement, Dpost
is a dummy that is set to 1
for the post-crisis sample of August 2009 to December 2017, Zi,t−1 contains a variety of ﬁrm-level
controls,20 αi is a ﬁrm ﬁxed eﬀect and αt is a time-ﬁxed eﬀect. The combination of the ﬁrm- and
time-ﬁxed eﬀect allows us to control for factors that are ﬁrm speciﬁc (but ﬁxed over time) and
aggregate patterns that aﬀect the level of the ﬁrm-speciﬁc implied volatility measure. Due to the
data availability of options data, our sample runs from January 1996 to December 2017.

t

The estimates are presented in Table D3 with two-way clustered standard errors along the ﬁrm
and time dimension. Column 1 does not include any ﬁrm-level controls, while column 2 adds the
full list of ﬁrm-speciﬁc controls listed above. For both columns, the coeﬃcient on leverage (δ)
is negative and signiﬁcant: In the pre-crisis sample ﬁrms with higher leverage had lower levels
of expected volatility on the day before the FOMC announcement. But the coeﬃcient on the
interaction of leverage and the post-crisis dummy (β) is positive and signiﬁcant: Relative to the
pre-crisis sample, leverage is more positively associated with implied volatility in the post-crisis
sample. In the post-crisis sample, expected volatility is roughly one-quarter of a standard deviation
higher for a ﬁrm that has leverage one standard deviation above the mean. Moreover, the size of
the total eﬀect in the post-crisis sample (δ + β) is positive and signiﬁcant, as shown by p-values in
the table. This means that as measured on the day before the FOMC announcement, high leverage
ﬁrms were expected to be less volatile in the pre-crisis sample but more volatile in the post-crisis
sample.

20The controls include the same as those in our baseline stock market regression, as well as the ﬁrm-level stock price

on the trading day prior to the FOMC day.

53

The results from Table D3 are also robust to using the alternative measures of leverage (debt-
to-assets and debt-to-equity), putting in time-sector ﬁxed eﬀects, excluding unscheduled FOMC
meetings from the sample, including the crisis dates in the post-crisis period or including ﬁnancial
ﬁrms in the sample. These results are discussed in Section 2.5.

In summary, the ﬁrm-level options data conﬁrm the reversal in the relationship between leverage
and monetary policy announcements since the ﬁnancial crisis. Moreover, our options-based variable
is measuring expected volatility as captured on the day before the FOMC announcement. The
change in the sign of the relationship between this expected volatility and leverage implies that
participants in the ﬁnancial markets have internalized the change in relationship since the crisis
that we estimated with stock returns in Section 2.3.1. Next, we provide evidence for this changing
relationship using ﬁrm-level investment data.

2.3.4 Evidence from investment data

In this section we corroborate the evidence from the stock market using ﬁrm-level variables on
economic activity from Compustat. Speciﬁcally we explore the response among S&P 500 ﬁrms
of ﬁrm-level investment and sales to monetary policy shocks. We focus in the main text on the
investment response, relegating our sales growth results to the Online Appendix. Our baseline
empirical speciﬁcation is the following:

∆ln(yit) = αi + αt +

n∈N

β1,nli,t−n−1m

t−n + β2,nli,t−n−1m

t−nDpost

t−n + Γ(cid:48)Zi,t−1 + eit

(2.4)

where yit is the value of either ﬁrm i’s real sales revenue or capital stock in quarter t, αi is a ﬁrm
i ﬁxed eﬀect, αt is a quarter t ﬁxed eﬀect, lit is ﬁrm i’s leverage ratio, m
is the sum of all high-
t
frequency monetary policy shocks that occur in quarter t, Dpost
is an indicator for the post-crisis
period, Zi,t−1 is a vector of ﬁrm-level controls (lagged by one quarter) and eit is the residual.21

t

21Zi,t−1 also contains each of the n lags of the ﬁrm’s leverage ratio and the respective interactions with the lagged
post-crisis dummy. As with the stock market speciﬁcation, the monetary policy shock m
is subsumed by the time
t
ﬁxed eﬀect. The same is true of the post-crisis dummy and its interaction with the monetary policy shock.

54

For the speciﬁcation with investment as the dependent variable, N = [0, 12]. Due to the strong
seasonality of sales data, N = {0, 4, 8, 12} for the speciﬁcation with sales growth as the dependent
variable.

The key parameters in the above speciﬁcation are β1,0 and β2,0, which estimate how the
responsiveness of the real variable, yit, to a contemporaneous quarterly monetary policy shock
diﬀers based on a ﬁrm’s leverage ratio in the pre-crisis and post-crisis, respectively. Since we
standardize lit to be mean zero and unit variance, these parameters can be interpreted as the
additional increase in a ﬁrm’s quarterly sales or investment in response to an expansionary monetary
shock by moving from an average leverage ratio to one standard deviation above the average leverage
ratio. As in the results discussed above, we multiply the monetary policy shock measure by negative
one so that an increase in m

t corresponds to an expansionary shock.

We control for factors in Zi,t−1 that could potentially aﬀect both a ﬁrm’s leverage ratio and
the quarterly growth in the ﬁrm’s sales or capital stock. For regressions with investment as the
dependent variable, Zi,t−1 includes the log of the book value of assets, the ratio of current assets
to total assets, the ratio of sales revenue to the (net) capital stock and the year-over-year real sales
growth. These controls are intended to capture the size of the ﬁrm and its liquidity/cash ﬂow, both
of which should increase a ﬁrm’s ability to ﬁnance investment expenditures. For regressions with
sales growth as the dependent variable, Zi,t−1 includes the log of the book value of assets and
the ratio of current assets to total assets. Since the monetary policy shocks occur throughout the
quarter and the ﬁrm-level variables are measured at the quarterly level, leverage and the controls
are lagged to ensure they are predetermined at the time of the monetary policy shocks. We also
include a dummy for the ﬁscal quarter in all speciﬁcations, to account for diﬀerences across ﬁrms
due to diﬀerent positions in their ﬁscal year. This is particularly important for sales, which displays
a strong seasonal trend based on ﬁscal quarter. Since we are using panel data, we are able to include
ﬁrm and time ﬁxed eﬀects. The ﬁrm ﬁxed eﬀect accounts for permanent characteristics of changes
in ﬁrm i’s sales or capital stock that are not captured by our controls. The time ﬁxed eﬀect accounts
for aggregate shocks common to all the ﬁrms in quarter t. The standard errors reported in the

55

parentheses are calculated using two-way clustering along the time and ﬁrm dimensions. Finally,
to ensure that the ﬁrm ﬁxed eﬀect is not endogenous, we only keep ﬁrms in the sample that have
at least 40 observations with non-missing sales or investment data in either the pre- or post-crisis
sample.

Table D4 shows the contemporaneous response of investment with the results for sales provided
in the Online Appendix. These results are consistent with the pattern emerging from the stock price
and implied volatility results. In the pre-crisis sample, the interaction between the monetary policy
shock and leverage is negative and signiﬁcant, while the interaction is positive and signiﬁcant
in the post-crisis sample. Our ﬁnding that investment is less responsive to a contemporaneous
monetary policy shock in the pre-crisis period matches the main ﬁnding of Ottonello & Winberry
(2018).22 Speciﬁcally, during the pre-crisis period, a ﬁrm with leverage one standard deviation
above average experiences an increase in investment 3.71% less than a ﬁrm with average leverage
during the quarter in which an expansionary monetary policy shock occurs.
In the post-crisis
period, a high-leverage ﬁrm would experience an increase in investment 7.86% more than a ﬁrm
with average leverage.

In section 2.5 below we discuss further robustness checks of these results, including using time-
sector ﬁxed eﬀects, adding more controls and expanding the sample beyond S&P 500 ﬁrms. Before
detailing these tests, we now turn to a discussion of potential mechanisms behind the empirical
ﬁndings we have presented up to this point.

2.4 Mechanism

We began by showing that there has been a change in the relationship between ﬁrm leverage and
monetary transmission. Speciﬁcally, using both high frequency stock market data and quarterly

22Jeenas (2018) ﬁnds that higher leverage ﬁrms become more responsive than lower leverage ﬁrms several quarters after
a monetary policy shock in the pre-crisis period. In contrast, Ottonello & Winberry (2018) do not ﬁnd a diﬀerential
eﬀect by leverage beyond the contemporaneous quarter. Ottonello & Winberry (2018) contains a lengthy discussion
of the methodological diﬀerences between these two papers. Due to limited amount of data in the post-crisis sample, it
is diﬃcult to discern any statistically signiﬁcant diﬀerences in the long-term response between the pre- and post-crisis
samples. We ﬁnd that the clearest change from the pre- to the post-crisis occurs in the contemporaneous sensitivity
of investment due to diﬀerences in ﬁrm leverage. Thus we choose to focus on these results here.

56

economic activity variables we showed that high leverage ﬁrms were less responsive in the pre-crisis
sample but more responsive in the post-crisis sample. In this section we shed some light on the
mechanism underlying the change in this relationship.

In the post-crisis sample, with the fed funds rate stuck at the zero lower bound, the Federal
Reserve has leaned more on unconventional monetary policy actions like large scale asset purchases
and forward guidance. While there is a large literature on the channels of unconventional policy
transmission, there is very little work that studies the heterogeneous transmission through the
balance sheets of non-ﬁnancial ﬁrms.
In this section we start with the recent heterogeneous
ﬁrm general equilibrium model of Ottonello & Winberry (2018) (OW hereafter) who study the
transmission of conventional monetary policy (in the pre-crisis sample) through ﬁrm balance sheets.
Within this framework, we discuss a channel that can rationalize our post-crisis ﬁndings and provide
supporting empirical evidence for it. Next, we dig deeper to understand this channel and show that
long rates have become more sensitive to monetary policy shocks in the post-crisis sample. Thus
a monetary policy induced change in long rates could have outsize spillover eﬀects on ﬁnancing
conditions for ﬁrms who rely more on longer term funding. Consistent with this story, we ﬁnd
that our baseline results are driven by ﬁrms whose leverage is more dependent on long-term debt.
Finally, we provide some suggestive evidence for an additional channel of monetary transmission to
ﬁrm balance sheets since the crisis, namely through changes in uncertainty about future monetary
policy decisions.

2.4.1 Ottonello & Winberry channels of monetary transmission

The canonical theoretical framework to understand the transmission of monetary policy through
ﬁrm balance sheets is the ﬁnancial accelerator model of Bernanke et al. (1999). The essential
feature of the models in this vein is the existence of some ﬁnancial friction in the borrower-lender
relationship. For our purposes, the key question is what this framework implies for the heterogenous
ﬁrm response to monetary policy. In the literature, the theoretical predictions of how ﬁrm balance
sheet characteristics aﬀect monetary transmission are ambiguous. Bernanke et al. (1999) did

57

extend their baseline model to a heterogenous (two-ﬁrm) case. With their preferred calibration
they ﬁnd that ﬁrms that have a larger external ﬁnance premium respond more strongly to monetary
policy shocks. Building on the work of Khan et al. (2016), OW extend the Bernanke et al. (1999)
framework to allow for a richer structure of heterogeneity (including ﬁrm-speciﬁc productivity and
capital quality shocks) and ﬁrm default. Contrary to Bernanke et al. (1999), they ﬁnd that ﬁrms with
higher leverage are less responsive to monetary policy. Moreover, they conﬁrm their results using
an empirical analysis for the pre-crisis sample. These OW results are consistent with our empirical
results shown above for the pre-crisis sample. If we start with the OW model as our baseline model
for the pre-crisis sample, is it possible to explain our post-crisis results in this framework? Below
we summarize their model in brief and layout the key mechanisms from their model to understand
this issue.

The OW model has ﬁrms that can invest in capital by borrowing or using internal funds and
generates default in equilibrium. They embed this heterogeneous ﬁrm setup into a standard New
Keynesian sticky-price framework to study the eﬀects of monetary policy. To understand the
relevant mechanism of monetary transmission, we reproduce a key ﬁrst-order condition from their
model. For a given level of productivity (z), the ﬁrst order condition for the optimal choice of a
ﬁrm’s investment (k(cid:48)) and borrowing (b(cid:48)) is given by23

(cid:18)

qt − εR,k(cid:48)(cid:0)z, k(cid:48)

, b(cid:48)(cid:1) b(cid:48)

k(cid:48)

(cid:19)

(z, k(cid:48), b(cid:48))

Rsp
t

1 − εR,b(cid:48) (z, k(cid:48), b(cid:48)) =

1
Rt

Et

(cid:2)MRPKt+1(cid:0)z(cid:48)
, k(cid:48)(cid:1)(cid:3)

The left hand side represents the marginal cost of capital and is a product of two terms. The
ﬁrst one is the price of capital net of the elasticity of the lender’s rate schedule with respect to
investment (εR,k(cid:48) (z, k(cid:48), b(cid:48))). An extra unit of investment costs qt but it adds to the ﬁrm’s collateral
23We have omitted two terms that capture the marginal beneﬁt of investment from this ﬁrst order con-
dition.
which is the covariance between
The second one is given by
the return to capital and the ﬁrm’s
1
Rt v0
∂k(cid:48)
and captures how more investment aﬀects a
t
ﬁrm’s default probability. OW ﬁnd that these two terms do not play a major role and we have thus omitted
them for convenience.

(cid:0)z(cid:48),k(cid:48),b(cid:48)(cid:1)(cid:1)
(cid:0)z(cid:48),k(cid:48)(cid:1),1+λt+1
Covt(cid:0)MRPKt+1
(cid:2)1+λt+1(z(cid:48),k(cid:48),b(cid:48))(cid:3)
(cid:0)k(cid:48),b(cid:48)(cid:1)
(cid:0)k(cid:48),b(cid:48)(cid:1)
(cid:19)

Et
shadow value of

1
Rt

g(cid:0)z(cid:0)k(cid:48), b(cid:48)(cid:1)(cid:1)(cid:18) ∂zt+1

(cid:0)k(cid:48), b(cid:48)(cid:1)(cid:17)

− ∂zt+1
∂b(cid:48)

The ﬁrst one is

(cid:16)

zt+1

resources.

58

and thus lowers the interest rate charged by lenders. The second term is how borrowing costs
(z, k(cid:48), b(cid:48)) is the ﬁrm-speciﬁc rate Rt(z, k(cid:48), b(cid:48)) (relative to the risk-free
change with investment. Rsp
t
rate Rt). This is scaled by one minus the elasticity of the debt price schedule (1 − εR,b(cid:48) (z, k(cid:48), b(cid:48)))
with respect to borrowing, which captures the idea that an increase in borrowing makes the ﬁrm
riskier and thus makes lenders charge a higher premium. Graphically (as can be seen in Figure E2),
the marginal cost schedule (as a function of capital accumulation) is ﬂat for low levels of capital
as the ﬁrm has enough cash on hand to not be perceived as risky. After a certain cutoﬀ point, the
marginal cost curve slopes upward as the higher level of borrowing required to fund the capital
increases the riskiness of ﬁrms. The right hand side represents the marginal revenue product of
capital discounted by the risk-free rate. Graphically, the marginal beneﬁt schedule is represented
by a standard downward sloping curve due to diminishing returns to capital.

What is the eﬀect of an expansionary monetary policy shock in this framework? By lowering
the risk-free rate, an expansionary shock lowers the discount rate and thus shifts the marginal beneﬁt
curve up and to the right.24 An expansionary shock has three eﬀects on the marginal cost curve.
First, it shifts up the curve because an increase in the demand for investment leads to an increase
in the price of capital. Next, this shock extends the ﬂat part of the marginal cost curve because it
increases the ﬁrm’s cash on hand and decreases the amount the ﬁrm needs to borrow to ﬁnance a
given amount of investment. Finally, it ﬂattens the upward sloping part of the curve because the
ﬁrm’s collateral is worth more and thus reduces the loss to the lender in case of default. These can
be seen in Figure E2.

How do ﬁrms with high and low leverage react diﬀerently to monetary policy shocks? In this
framework there are competing channels which make it theoretically ambiguous whether a high
or low leverage ﬁrm will respond more. For a high-leverage ﬁrm, the upward sloping part of the
marginal cost curve is steeper and thus this will make it less responsive to monetary policy induced
shifts of the marginal beneﬁt curve. On the other hand, a high leverage ﬁrm’s marginal cost curve
will ﬂatten more in response to an expansionary monetary shock, making it more responsive. In

24There are also general equilibrium eﬀects due to changes in the price of output, capital and wages which in the OW

calibration further shift out the marginal beneﬁt curve.

59

the OW calibration they ﬁnd that the former eﬀect dominates and thus a high leverage ﬁrm is less
responsive to monetary policy shocks. This case is highlighted in the top row of Figure E2. So
how can we explain our results of higher sensitivity for high leverage ﬁrms in the post-crisis sample
using this framework?

Theoretically, there are three possible ways in which this can happen. In the post-crisis sample
we would need that i)the marginal beneﬁt curve shifts more for high leverage ﬁrms in response to a
monetary shock or ii)the slope of the marginal cost curve is more ﬂat (on average, not in response
to monetary shocks) for high leverage ﬁrms (relative to low leverage ﬁrms) or iii) the slope of the
marginal cost curve ﬂattens more in response to a monetary shock for high leverage ﬁrms and that
this increased ﬂattening is enough to outweigh the relative steepness of high leverage ﬁrms.

We argue that the ﬁrst two explanations are less plausible and provide evidence that the third
explanation is likely at play. Regarding the ﬁrst explanation, the shift of the marginal beneﬁt curve
is driven by changes in the discount rate. It is unlikely that discount rates for high leverage ﬁrms
respond diﬀerentially in the post-crisis samples.25 The second explanation would require that in
the post-crisis sample high leverage ﬁrms are perceived to be less risky than low-leverage ﬁrms. In
other words, the credit spread charged by lenders (relative to the risk-free rate) to high leverage ﬁrms
would increase less as these ﬁrms take on more borrowing. First, recall that in FigureE1 above we
have shown that a ﬁrm’s leverage position is fairly stable across the two samples. Moreover, Figure
E1 shows that the correlation of leverage with measures of ﬁrm riskiness are stable across the pre-
and post-crisis samples. This rules out the unlikely scenario that our results are being driven by the
high leverage ﬁrms somehow becoming less risky in the post-crisis sample.

This leaves us with the third explanation. This requires that an expansionary monetary policy
shock would ﬂatten the marginal cost curve of high leverage ﬁrms more (relative to low leverage
ﬁrms). Additionally this increased ﬂattening would have to be large enough to overcome the relative
steepness of the marginal cost curve for high leverage ﬁrms. From the ﬁrst-order condition above,
the marginal cost curve ﬂattening more would imply that the credit spread charged by the lender

25There are also general equilibrium eﬀects that work through the price of output goods, capital and wages but these

are also unlikely to respond diﬀerentially for high leverage ﬁrms in the post-crisis sample.

60

(relative to the risk-free rate) to a high leverage ﬁrm falls more (relative to a low leverage ﬁrm) in
response to an expansionary monetary shock. We can see this readily from the bottom row of Figure
E2. This ﬁgure shows in the post-crisis sample that even though the slope of a high leverage ﬁrm
is unconditionally steeper than a low leverage ﬁrm, it ﬂattens more in response to an expansionary
monetary shock to make the desired change in investment higher for high leverage ﬁrms.

This explanation provides a simple testable implication: the credit spread (relative to the risk-
free rate) of high leverage ﬁrms should fall more in response to an expansionary monetary shock
in the post-crisis sample. While we do not have access to high-frequency ﬁrm-level bond yields,
we use bond indices that group together ﬁrms with similar risk proﬁles. Speciﬁcally, we use the
Moody’s bond yields index on ﬁrms rated AAA and those rated BAA.26 Recall that Figure E1
showed that there is a negative and signiﬁcant correlation between leverage and credit rating in both
the pre- and post-crisis samples. Our hypothesis then is that the spread between the BAA yield and
risk free rate falls more in the post-crisis sample relative to the spread between the AAA yield and
the risk-free rate. Or alternatively, the BAA-AAA yield spread should fall more in the post-crisis
sample. We test the hypothesis using the following regression

∆ln(yt) = α0 + α1Dpost

t

+ δεm

t + βDpost

t

εm
t + eit

(2.5)

t

where yt is the Moody’s BAA-AAA bond yield spread, Dpost
is a dummy for the post-crisis period
and εm
is the monetary policy shock. The results presented in Table D5 show that indeed this is
t
the case. In the pre-crisis sample, this bond spread response to monetary policy is small and not
statistically signiﬁcant. However, in the post-crisis sample the bond spread response is negative
and signiﬁcant at the 10% level. In response to an expansionary monetary policy shock that lowers
the two year rate by 1%, the BAA-AAA bond spread falls 20% more in the post-crisis sample. We
interpret this as suggestive evidence that monetary policy shocks are ﬂattening the marginal cost
curve more for ﬁrms with higher leverage in the post-crisis sample, thus driving their increased
responsiveness. Overall, this result is consistent with the work of Gilchrist & Zakrajšek (2013) who

26Over 70% of the ﬁrms in our sample are rated BAA or higher.

61

ﬁnd using credit default swap data that QE announcements reduced the cost of insuring against
default risk more for higher risk ﬁrms.

2.4.2 Transmission through changes in long-term funding conditions

Why is it that monetary policy actions ﬂatten the marginal cost more and thus compress the
yield spread between high leverage and low leverage ﬁrms more in the post-crisis sample? The
explanation that we pursue in this subsection relies on the fact that long rates have become more
sensitive to monetary policy in the post-crisis sample and thus monetary shocks can potentially
aﬀect ﬁnancing conditions more for ﬁrms that rely on long-term funding.

It is well established in the literature that monetary policy has substantial eﬀects on long-term
rates. Gürkaynak et al. (2005) showed this for nominal rates and more recently Hanson & Stein
(2015a) showed this even for real rates. We ﬁrst document that both nominal and real long-term
rates (and term premium estimates commonly used in the literature) have become more sensitive
to monetary policy shocks in the post-crisis sample.

In Table D6 we regress the (daily) change in the 10 year nominal yield, the 10 year real yield
and the term premium of the 10 year nominal yield on the monetary policy shock. The nominal
yields are from Gürkaynak et al. (2007), the real yields from Gürkaynak et al. (2010) and the
term premium estimates are from Kim & Wright (2005). The table shows our baseline measure
(MP shock) in Panel A and the 2 year shock in Panel B. An expansionary 2 year shock lowers all
three: nominal yield, real yield and the term premium of the 10 year bond. Importantly, the fall in
these three variables is substantially larger in the post-crisis sample.27 The diﬀerence between the
pre- and post-crisis coeﬃcients is also statistically signiﬁcant for the real yield and term premium
response. In response to a 100 basis point reduction in the 2 year rate, the real yield (term premium)
falls by 33 (21) basis points in the pre-crisis sample but by 101 (63) basis points in the post-crisis
sample. Thus the real yield and the term premium on the 10 year Treasury bond are three times
as sensitive to monetary policy shocks in the post-crisis sample relative to the pre-crisis sample.

27These results are consistent with the pattern documented in Hanson et al. (2017).

62

Recall that our baseline measure combines the fed funds rate target shocks with the 10 year shocks;
thus, it is not straightforward to interpret the regressions reported in Panel A. But, we include them
for completeness and moreover these results conﬁrm the ﬁndings from the 2 year rate regressions.
These results suggest that monetary policy may be having an outsize eﬀect on long-term
funding conditions in the post-crisis sample. This could translate into a bigger eﬀect on ﬁrms that
are more reliant on long-term funding, for example through the “gap-ﬁlling" framework outlined in
Greenwood et al. (2010). Foley-Fisher et al. (2016) ﬁnd some supporting evidence of this channel
by studying the Federal Reserve’s Maturity Extension Program (MEP) in 2011 and 2012. They
ﬁnd that MEP had disproportionate eﬀects on ﬁrms that had more long-term debt. The relevant
question for us is whether ﬁrms whose leverage is driven more by long-term debt are driving our
results regarding the changing relationship since the crisis. To test this mechanism we focus on
a long-term component of our overall leverage measure. Our baseline leverage measure (debt-to-
debt + equity where LT
capital) is deﬁned as leverage =
debt is deﬁned as all debt maturing in more than one year. We separate ﬁrms as having high LT
leverage (top two-thirds of the LT leverage distribution) and low LT leverage (bottom third of the
LT leverage distribution). Then we run our baseline regressions separately for these two groups in
the pre- and post-crisis samples.

debt + equity. We now deﬁne LT leverage =

debt

LT debt

The results are presented in Table D7. The ﬁrst two columns (1a and 1b) show the baseline
regressions for the ﬁrms classiﬁed as “High LT leverage" and the next two columns (2a and 2b)
show them for the “Low LT Leverage" ﬁrms. For the high LT leverage ﬁrms, we get the same
ﬂip in the sign of the interaction coeﬃcient as the baseline results, with a negative and signiﬁcant
coeﬃcient in the pre-crisis sample but a positive and signiﬁcant coeﬃcient in the post-crisis sample.
For the low LT leverage ﬁrms, the coeﬃcient is negative and signiﬁcant in the pre-crisis sample
but insigniﬁcant and essentially zero in the post-crisis sample. Thus, in the post-crisis sample,
our baseline results of high leverage ﬁrms being more responsive to monetary policy shocks is
driven by ﬁrms whose leverage is more dependent on long-term debt. This is consistent with the
interpretation that in the post-crisis sample monetary policy aﬀects ﬁrms that are more exposed

63

to funding using long-term instruments. Moreover, these results are consistent with the work of
Foley-Fisher et al. (2016) who studied the MEP program. We have run our results dropping the two
MEP related FOMC meetings in 2011 and 2012 and ﬁnd that they are essentially identical to those
reported in Table D7. Thus our results indicate that the phenomenon of monetary policy having
a bigger eﬀect on ﬁrms that are more exposed to long-term debt has been true more generally of
Federal Reserve policy since the crisis and not just speciﬁc to the MEP.

2.4.3 Transmission through monetary policy uncertainty changes

In the previous subsection we provided some evidence that ﬁrms with leverage more dependent on
long-term debt were more responsive to monetary policy shocks in the post-crisis sample. Here
we explore another channel through which monetary policy actions since the crisis could have had
heterogenous eﬀects, namely the uncertainty channel.

There is a growing literature that highlights the role of changes in policy uncertainty as a channel
of monetary transmission.28 Moreover, Bauer et al. (2019) and Kroencke et al. (2018) highlight the
growing importance of this uncertainty channel in the post-crisis sample. To investigate the role
of this channel we use the high-frequency market-based measure of monetary policy uncertainty
developed by Bauer et al. (2019). They use options on Eurodollar futures to get a model-free
estimate of the standard deviation of the expected interest rate at horizons of 0.5 year, 1 year, 1.5
years and 2 years. We construct a single series by taking the ﬁrst principal component of these four
measures.29 We will use the (daily) change in this principal component measure on FOMC days as
the monetary policy uncertainty shock (labeled “MPU shock").

First, we explore the eﬀect of this monetary policy uncertainty measure on ﬁrm-level stock
returns and whether high leverage ﬁrms respond diﬀerentially. We estimate the following regression

mpu
si,t = αi + αt + βli,t−1
t

+ δli,t−1 + Γ(cid:48)Zi,t−1 + ei,t

(2.6)

28See the recent work of Creal & Wu (2017), Husted et al. (2017), Bauer et al. (2019), Kroencke et al. (2018) and

Bundick et al. (2017b)

29The estimates for the 1.5 year and 2 year horizons are not available for the entirety of our sample. Thus, we perform

separate principal component analyses on three subsamples, partitioned by the number of measures available.

64

where we have replaced the monetary policy shock from our baseline speciﬁcation in Equation
mpu
2.2 with the MPU shock (
). Because of the limited availability of the uncertainty measure,
t
our sample starts in January 1994. The MPU shock is scaled so that a positive number reﬂects a
lowering of uncertainty (“an expansionary shock"). Table D8 shows the regression results for the
pre- and post-crisis sample. In the pre-crisis sample the interaction coeﬃcient of leverage and MPU
shock is negative but small in magnitude and statistically insigniﬁcant: in the pre-crisis sample,
leverage did not play a role in ﬁrms’ stock price response to uncertainty shocks. However, this
coeﬃcient is positive and statistically signiﬁcant in the post-crisis sample: in the post-crisis sample
ﬁrms with higher leverage are more responsive to monetary policy uncertainty shocks.30

We also ﬁnd some additional evidence for the heterogenous eﬀects of the monetary policy
uncertainty shocks using ﬁrm-level implied volatility constructed from options data. We ﬁnd that
the change in the expected volatility of ﬁrm-level stock returns responds signiﬁcantly to monetary
uncertainty shocks. Moreover, this transmission is stronger for high leverage ﬁrms in the post-crisis
sample. Speciﬁcally, we estimate the following regression

mpu
∆ivoli,t = αi + αt + β1li,t−1
t

+ β2Dpost

t

mpu
li,t−1
t

+ δli,t−1 + Γ(cid:48)Zi,t−1 + ei,t

(2.7)

t

where ∆ivoli,t is the (daily) change in ﬁrm-level implied volatility on FOMC days, αi is a ﬁrm ﬁxed-
mpu
eﬀect, αt is an FOMC day ﬁxed-eﬀect, li,t−1 is leverage, 
is the monetary policy uncertainty
t
shock, Dpost
is a dummy for the post-crisis period and Zi,t−1 is a vector of ﬁrm-level controls
containing the ﬁrm’s share price change, real sales growth, size, price-to-cost margin, receivables-
minus-payables to sales, depreciation to assets, age, log(market cap), current assets to total assets
and an indicator for current ﬁscal quarter. Table D9 shows that β1 is positive and signiﬁcant.
Since the baseline eﬀect of a reduction in uncertainty is to reduce expected volatility, a positive
β1 implies that expected volatility fell less on FOMC days in response to reduced uncertainty for
ﬁrms with high leverage in the pre-crisis sample. In other words, in the pre-crisis sample an FOMC
announcement that lowered uncertainty would translate into a smaller reduction in expected future
30One issue with this uncertainty measure is that it is only available at a daily frequency. Since our monetary policy
shock is constructed at a higher frequency, it is not straightforward to compare the relative contribution of our baseline
monetary policy shocks versus the monetary policy uncertainty shocks.

65

volatility for a ﬁrm with high leverage. But the estimate of β2 is negative and signiﬁcant. This
suggests that this relationship changes after the crisis. In the post-crisis sample, the response of a
ﬁrm with high leverage to uncertainty shocks is more negative, i.e. high leverage ﬁrms are more
responsive to reductions in uncertainty.

2.5 Robustness Checks

We start by documenting the robustness of our three diﬀerent empirical approaches to using an
alternative deﬁnition of leverage: debt-to-assets. Debt-to-assets is widely used in the literature to
measure ﬁrm leverage, e.g. Whited & Wu (2006) ﬁnd that Compustat ﬁrms with a higher debt-to-
assets ratio are more ﬁnancially constrained. Table D14 shows that these results using debt-to-assets
conﬁrm the baseline results. Our baseline results use the four-quarter moving average of debt-to-
capital31; however, one could be concerned that this smooths out meaningful, higher-frequency
variations in leverage. In Table D15, we show that our results are robust to using the one-quarter
lagged version of our leverage measure.

Next, we tackle the concern that our results may be driven by diﬀerent sectors being more or
less responsive to monetary policy shocks. To account for this we include a sector by FOMC day
ﬁxed eﬀect, rather than just an FOMC day ﬁxed eﬀect. In Table D16, we show that the signiﬁcance
and magnitude of our baseline results are not meaningfully aﬀected.

As a further robustness check, we expand our sample beyond S&P 500 ﬁrms. One main
diﬀerence between the S&P 500 sample and a full CRSP/Compustat sample is that ﬁrms in the
expanded sample enter and exit much more frequently, as well as have more missing data values.
To limit the eﬀects of changes in sample composition amongst the full CRSP/Compustat sample,
we keep only those ﬁrms that are present in our entire sample period. These results are displayed
in Table D17. We see that our baseline results are robust to broadening our sample to include more
than just S&P 500 ﬁrms.32

31To be consistent with Ottonello & Winberry (2018), our baseline investment table uses the one-quarter lagged debt-
to-capital ratio. We show in the Online Appendix, as well as Table D16, that our investment results become stronger
in terms of magnitude and signiﬁcance when using the four-quarter moving average of the debt-to-capital ratio.

32Due to the limited availability of expected volatility data for ﬁrms outside the S&P 500, we include only our stock

66

The Online Appendix contains several more robustness checks. First, we present the results
when we include the crisis dates in the post-crisis sample. The coeﬃcients show that including
the ﬁnancial crisis dates does not materially change the results. Second, we show that our stock
market results are not driven by a change in the composition of the sample between the pre-crisis
and post-crisis periods. In the Online Appendix we rerun our baseline stock market speciﬁcation,
limiting the sample to only those ﬁrms with non-missing data for all 221 FOMC days in our sample.
Despite losing approximately 75% of our sample to this restriction, the results qualitatively match
our baseline results: higher leverage ﬁrms are less responsive in the pre-crisis period and more
responsive in the post-crisis period. This shows that our main results are not caused by certain
ﬁrms entering or exiting the sample, e.g. ﬁrms that did not survive the ﬁnancial crisis.

FOMC meetings that are unscheduled can have eﬀects on ﬁnancial markets that are diﬀerent
from regularly scheduled meetings as the unscheduled meetings typically occur in times of economic
turmoil. The unscheduled meetings are also instances in which the Federal Reserve is more likely to
release information about economic fundamentals, see for example Lakdawala & Schaﬀer (2019a).
Thus we want to make sure that our results are not driven by these unscheduled meetings. This
issue only arises in the pre-crisis sample which has 16 unscheduled meetings, while our post-crisis
sample has none. The Online Appendix shows the regression results excluding the unscheduled
meetings. These results are quite similar to the baseline case. There may still be a concern that
even on regularly scheduled FOMC meetings the high frequency monetary policy shocks contain
a substantial information component. To address this concern we use forecast data (following the
approach in Lakdawala (forthcoming)) to cleanse the monetary policy shock of any information
eﬀects. These results, also shown in the Online Appendix, conﬁrm that our baseline results are not
driven by this issue. Finally, we show in the Online Appendix that including ﬁnancial ﬁrms in our
sample does not aﬀect our results.

market and investment results.

67

2.6 Conclusion

In this paper we add to the growing empirical literature on monetary policy and ﬁrm-level
heterogeneity. Using both high frequency data from the stock market and lower frequency invest-
ment data we show that the role of leverage in transmitting monetary policy shocks has changed
since the ﬁnancial crisis. Before the ﬁnancial crisis a ﬁrm with higher than average leverage was
less responsive to monetary policy shocks. However, after the ﬁnancial crisis this relationship has
reversed so that ﬁrms with higher leverage are now more responsive to monetary policy shocks.
We interpret our pre-crisis results through a structural model and provide suggestive evidence for
a mechanism that can rationalize our post-crisis results. Since the crisis, long rates have become
more sensitive to monetary policy shocks suggesting increased sensitivity of long-term funding
conditions to monetary policy. Consistent with this story, we show that our baseline results of
increased responsiveness since the crisis are driven by ﬁrms whose leverage is more dependent
on long-term debt. Finally, we also provide some suggestive evidence that monetary transmission
related to leverage works in part through an uncertainty channel, where FOMC announcements
change the market’s perceived uncertainty about future policy actions.

Our results have potentially important implications for the aggregate eﬀects of monetary policy.
Focusing on the pre-crisis sample, Ottonello & Winberry (2018) ﬁnd that monetary policy is
less eﬀective in the aggregate when there is a bigger share of riskier ﬁrms in the economy. Our
estimates from the post-crisis sample indicate that this relationship has reversed in the last decade.
This suggests two important avenues for future research. First developing general equilibrium
models with ﬁrm heterogeneity that also allow a role for unconventional monetary policy will help
us understand the aggregate transmission of monetary policy since the crisis. Second, we think that
further exploring empirical strategies to tease out the state-dependent eﬀects of monetary policy
based on ﬁrm balance sheets is a promising area for future research.

68

CHAPTER 3

THE INTERNATIONAL SPILLOVER EFFECTS OF US MONETARY POLICY

UNCERTAINTY

3.1 Introduction

Recent work has highlighted the phenomenon of the global ﬁnancial cycle and the crucial
driving role of US monetary policy (Miranda-Agrippino & Rey (2015), Jordà et al. (2019)).
The literature has documented a variety of transmission channels through which Federal Reserve
actions aﬀect international ﬁnancial markets. However, most of the focus in the literature has been
on transmission through changes in the level of the Federal Reserve’s policy rate (i.e. ﬁrst moment
changes).1 In this paper we show that changes in uncertainty around the level of the policy rate
(i.e. second moment changes) constitute an important additional dimension through which Federal
Reserve decisions transmit to international ﬁnancial markets.

The importance of uncertainty as an additional dimension of FOMC announcements is readily
seen from the Aug 2011 FOMC meeting. At this meeting the FOMC introduced explicit calendar-
based forward guidance for the ﬁrst time by saying that rates will be low until “. . . mid-2013".
Commonly used measures of the ﬁrst-moment monetary policy shocks (e.g. changes in futures
rates up to 1 year ahead) did not move much in response to the announcement but the market-
perceived uncertainty fell substantially. In this paper we conduct a systematic evaluation of how
such FOMC-induced changes in uncertainty transmit to international ﬁnancial markets.

To perform this analysis, we use an event-study framework around FOMC meetings with bond
yield data for a panel of 31 advanced and emerging countries.2 Our measure of monetary policy
uncertainty is the recent one developed by Bauer et al. (2019). This measure relies on high frequency

*This chapter is joint work with Aeimit Lakdawala and Matthew Schaﬀer.
1Albagli et al. (2019), Gilchrist et al. (2019) and Curcuru et al. (2018) are some recent examples from this literature.
There are also numerous papers that study the transmission of unconventional monetary policy actions and we discuss
how our work relates to that literature below.
2While the primary focus is the response of global bond yields, we also show that global equity markets react
considerably to monetary policy uncertainty.

69

options data to calculate the market’s perceived uncertainty; the conditional, risk-neutral 1 year
ahead standard deviation of changes in the Federal Reserve’s policy rate. We ﬁnd that an increase
in this market-based uncertainty raises bond yields in both advanced and emerging countries. This
eﬀect is over and above the eﬀect of surprise changes in the expected policy rate, which is the most
widely used measure of monetary shocks. In other words, uncertainty matters even after controlling
for ﬁrst moment shocks. While the average eﬀect is moderate, we document that the international
spillover through monetary policy uncertainty is larger when the Federal Reserve made deliberate
changes to its forward guidance language in the FOMC statement. This suggests that by changing
its communication about the uncertainty of its future actions, the Federal Reserve has an additional
policy tool through which it can aﬀect international ﬁnancial conditions.3

What is the channel through which changes in US monetary policy uncertainty aﬀect inter-
national bond yields? We ﬁnd that there is a key diﬀerence in the mechanism through which
monetary policy uncertainty is transmitted to yields in advanced versus emerging countries. In
advanced countries, the response works through an international portfolio balance channel whereas
for emerging countries it is related to a ﬂight to safety channel. We perform a variety of analyses
to better understand the diﬀerential transmission mechanisms.

First, we note that a standard asset pricing framework implies that the excess return of a long-
term bond is a function of its conditional volatility (among other things). A change in the market’s
perceived uncertainty about future short rates is a clear signal about long-term bonds’ conditional
volatility. Thus an increase in uncertainty would imply that investors should demand a greater
premium for holding long-term bonds. We document that changes in monetary policy uncertainty
do indeed aﬀect US bond yields through risk/term premia. This result is consistent with the recent
work of Bundick et al. (2017a) and Bauer et al. (2019).

Next, we decompose changes in international bond yields into an expected (or risk-neutral)
component and a term premium component. This is done using the methodology of Joslin et al.

3From a structural perspective, monetary policy uncertainty shocks can arise from variance of the residual of a policy
rule or from uncertainty about the reaction function part of the policy rule. See Appendix Section F.0.1 for a detailed
discussion.

70

(2011) and applying the bias correction of Bauer et al. (2012).
Interestingly, we ﬁnd that US
monetary policy uncertainty aﬀects bond yields in advanced countries only through the response of
the term premium whereas the emerging country response is entirely due to changes in the expected
component.

Taking up the advanced country response ﬁrst, we show that the eﬀect of monetary policy
uncertainty on term premia in these countries works through changes in the term premium on the
US 10 year Treasury bond.4 Thus, the transmission channel runs from monetary policy uncertainty
to US term premia and eventually to advanced country term premia. This mechanism is consistent
with recent theoretical work on the so-called international portfolio balance channel, for example
see Alpanda & Kabaca (2019). In this framework investors view bonds in diﬀerent countries as
imperfect substitutes for each other and this creates a link for changes in bond term premia in
one country to aﬀect those in another.5 A key implication from this framework is that the size
of this spillover should directly depend on the degree of substitutability between the countries’
bonds. We provide a test of this hypothesis by constructing an empirical measure of the degree of
substitutability between the bonds of a foreign country with the US. This measure is similar to the
recent work of de los Rios & Shamloo (2017) and related to the older idea of Frankel (1982). As
expected, we ﬁnd that bonds of advanced countries are viewed as more substitutable with the US.
Importantly, and consistent with the theoretical framework, we ﬁnd that the term premium response
to uncertainty in a given country is larger if its bonds are more substitutable with the US.

These results also shed light on the driving force behind some recent empirical work in the
literature on the international spillover of US ﬁnancial markets. Two recent papers (Mehrotra
et al. (2019) and Curcuru et al. (2018)) ﬁnd that changes in US term premia have a stronger eﬀect
on yields in advanced countries relative to emerging countries, consistent with our results. Our

4Speciﬁcally, in a regression of advanced country term premia on uncertainty, the coeﬃcient on uncertainty is positive
and signiﬁcant. But when we control for the US 10 year term premium in this regression, the coeﬃcient on uncertainty
essentially goes to zero.
5The theoretical papers in this ﬁeld do not explicitly focus on monetary policy uncertainty as the originating source for
the transmission. However, the transmission mechanism is relevant as long as there is an eﬀect through term premia.
For example, Alpanda & Kabaca (2019) focus on the role of quantitative easing in driving term premia, while we are
interpreting this eﬀect coming from changes in monetary policy uncertainty.

71

work highlights that an important source of this transmission through term premia is driven by
FOMC-induced changes in uncertainty.

US monetary policy uncertainty transmission to emerging countries works through a diﬀerent
channel than the one we highlighted for advanced countries. Changes in monetary policy uncertainty
show up not in the term premium component of emerging country bonds, but rather the expected
component. In other words, after an increase in uncertainty, markets expect that interest rates in
emerging countries will rise. Using a local projections framework we conﬁrm that the market
reaction is correct; short rates in emerging countries do indeed rise and are higher a year after the
shock. This response in emerging countries is tied to a ﬂight to safety channel whereby an increase
in uncertainty makes investors pull capital out of countries that are perceived to be risky. Using
the Treasury’s TIC monthly capital ﬂows, we show that net holdings of emerging country (but not
of advanced country) bonds decrease in response to monetary policy uncertainty shocks. Overall
these results are consistent with the recent idea that capital ﬂows in emerging countries are quite
risk-sensitive (Kalemli-Ozcan (2019)).

The recent work of Rey (2013) suggests that the classic Mundell-Fleming “trilemma" may have
morphed into a “dilemma", whereby ﬂexible exchange rates do not insulate countries from ﬁnancial
spillovers unless there are additional restrictions on capital mobility. Kalemli-Ozcan (2019) argues
that for emerging countries this is not quite the case, at least in response to ﬁrst moment US monetary
shocks. We ﬁnd that the emerging country response to US monetary uncertainty is not related to
the exchange rate regime, but rather to that country’s ﬁnancial openness. Speciﬁcally, the higher
is the Chinn & Ito (2006) index of capital account openness, the larger is that country’s response
to uncertainty. We also investigate more broadly if there is heterogeneity across countries in their
responsiveness to monetary policy uncertainty following the methodology of Iacoviello & Navarro
(2019). For advanced countries, we do not ﬁnd evidence of this. The usual country characteristics
that are used in the literature to explain cross-country diﬀerences do not matter for reaction to
monetary policy uncertainty.6 For emerging countries we see somewhat more heterogeneity across

6Our baseline variables include ﬁnancial depth, exchange rate regime, trade openness, capital account openness and
the short-term interest rate diﬀerential with the US. We also attempted speciﬁcations using trade with the US and

72

countries but do not ﬁnd much other than ﬁnancial openness that can explain the responsiveness of
yields.7

There are diﬀerences in the advanced and emerging country yield response over time. The
advanced country response is relatively stable over our full sample from January 1995 to June
2019. But for emerging country bonds the responsiveness to monetary policy uncertainty is only
prevalent in the period since the ﬁnancial crisis. This suggests that in more recent years not only has
the interconnectedness of global ﬁnancial markets increased but shocks originating in one country
are now being transmitted through new channels. These results highlight the need for exploring
and incorporating these uncertainty-based transmission mechanisms in theoretical open economy
macroeconomic models.

We also investigate the impact of US monetary policy uncertainty on international stock prices.
An increase in US monetary policy uncertainty leads to a reduction in stock prices in both advanced
and emerging countries, but only in the period since the ﬁnancial crisis.8 For both advanced and
emerging countries, stock markets respond more to uncertainty in the post-crisis sample.

For long term yields and equity prices in both advanced and emerging countries, the size of the
response to monetary policy uncertainty is larger than the response to the conventional ﬁrst moment
policy surprises.9 Moreover, accounting for changes in the second moment is important even if
one is only interested in the ﬁrst moment eﬀect of international spillovers. Leaving out changes in
monetary policy uncertainty in the event-study regression biases the estimated eﬀect of monetary
policy surprises. This is because there is a positive correlation between changes in the ﬁrst and
second moment. Our estimates suggest that omitting uncertainty can lead to overstating the eﬀects
of monetary policy surprises by up to 50%.10 Moreover, we document substantial increases in R2

dollar debt exposure.
7Consistent with our results, Bowman et al. (2015) ﬁnd that ﬁnancial vulnerability (including capital account openness)
is one of the main determinants of yield responses.
8This result conﬁrms a general pattern that is consistent with the investigation of Indian stock markets carried out
in Lakdawala (2018), where they attribute the increased responsiveness to the growing role of foreign institutional
investors in domestic equity markets.
9For shorter maturity yields, relative to the ﬁrst moment policy surprise, the uncertainty eﬀect is somewhat smaller in
advanced countries and roughly the same size in emerging countries.
10Our uncertainty measure is a risk-neutral measure and thus (as we explain below) the 50% estimate may be better

viewed as an upper bound.

73

when adding uncertainty to the event-study regressions. Thus we argue that leaving out monetary
policy uncertainty gives an incomplete picture of the transmission of Federal Reserve actions to
international ﬁnancial markets.

Our paper builds on the work of Bauer et al. (2019) that develops the measure of monetary
policy uncertainty used in our paper. The main focus in Bauer et al. (2019) is on exploring how
uncertainty is related to forward guidance language used by the FOMC and its transmission to
domestic ﬁnancial markets in the US. In this paper we focus on the international transmission of
changes in uncertainty to bond and stock markets in a large set of advanced and emerging countries.
There is also a substantial literature that focuses on the international eﬀects of unconventional
Federal Reserve actions, e.g.
see Anaya et al. (2017), Bhattarai et al. (2015), Bowman et al.
(2015), Neely (2015), Fratzscher et al. (2018), and Kolasa & Wesołowski (2020). We show that
the transmission of uncertainty is not altered on FOMC meeting dates with notable quantitative
easing announcements. There is a view that attributes quantitative easing announcements as
working through a signal about the future expected path of the policy rate, the so-called signalling
channel (Bauer & Rudebusch (2014) and Bauer & Neely (2014)). Our results suggest that signals
about the uncertainty around future rates are an additional and largely unexplored dimension of
unconventional monetary policy transmission.

In addition to the literature that studies the eﬀects of US monetary policy on asset prices (e.g.
Albagli et al. (2019), Gilchrist et al. (2019), Curcuru et al. (2018), Ehrmann et al. (2011) and
Hausman & Wongswan (2011)), there is also a large literature exploring the eﬀects on capital ﬂows
(e.g. Kalemli-Ozcan (2019), Dahlhaus & Vasishtha (2014), Chari et al. (2020)). But this literature
estimates the eﬀect of changes in the level of the Federal Reserve’s policy rate. We extend this
literature to consider the international spillovers of changes in US monetary policy uncertainty.

While there is also a growing literature studying the international spillover eﬀects of overall US
uncertainty (see for example Bhattarai et al. (2019) and Carrière-Swallow & Céspedes (2013)), few
papers have explored the international transmission of monetary policy speciﬁc uncertainty. On
the empirical front, Lakdawala (2018) shows that the eﬀect of US monetary policy uncertainty on

74

the Indian stock market has grown since the ﬁnancial crisis. Gupta et al. (2020) study uncertainty
spillovers between a sample of nine advanced countries using the measure of Istreﬁ & Mouabbi
(2018), which primarily captures disagreement among professional forecasters. A related theoret-
ical work is Ghironi & Ozhan (2019), which investigates the impact of shocks to the variance of
the domestic country’s policy rate. Our emerging country results are broadly consistent with their
framework.

3.2 US Monetary Policy Shocks

An increasingly common approach in the literature to measure US monetary policy shocks is
to use changes in futures rates around FOMC announcements.11 This measure is a ﬁrst moment
shock that captures surprise changes to the expected path of the FOMC’s policy rate. The main
contribution of this paper is to show that a second moment shock (i.e surprise changes to the
uncertainty around the expected path of the FOMC’s policy rate) also has substantial spillover to
international ﬁnancial markets. In Appendix Section F.0.1 we frame monetary policy uncertainty
through the lens of a simple structural monetary policy rule. There we show that monetary policy
uncertainty can come from variance of the residual in the monetary policy rule and also uncertainty
about the reaction function part of the policy rule. Next, we detail the construction of this shock
from option prices and provide a discussion of how prominent changes in our measure are related
to speciﬁc changes in the forward guidance language used by the FOMC. While the focus will be
on the transmission of this uncertainty measure, we also include the traditional ﬁrst moment shock
because, as we discuss below, the two measures are correlated.12

3.2.1 Monetary Policy Uncertainty

To construct the monetary policy uncertainty (mpu) measure, we use the methodology of Bauer
et al. (2019). The object of interest is the standard deviation of the federal funds rate τ-periods ahead

11For example see the early work of Kuttner (2001). For more recent work see Nakamura & Steinsson (2018).
12We view our approach of separately studying the second moment shock transmission as complimentary to the large

literature that studies the overall eﬀect of both conventional and unconventional monetary policy actions.

75

conditional on the current information at time t, i.e.(cid:112)Var(FFRt+τ|It). The methodology provides

a model-free estimate of the risk-neutral conditional standard deviation, given prices of futures and
options at time t. Our baseline measure will set τ to 12 months to measure the uncertainty about
the 1 year ahead rate. The change in this measure is calculated in a two-day window around the
FOMC announcement. We scale our measure to have unit standard deviation.

We refer the reader to Appendix Section F.0.2 for the details of the construction of the uncertainty
measure. Here we provide a brief discussion of the relevant empirical properties and also what
drives the big changes in our measure. Figure H1 plots our baseline measure:
the change in
the standard deviation of the 1 year ahead expected rate in a two day window around the FOMC
announcement. We label this measure mpu in the regression analysis that follows. On average,
our measure declines on FOMC days: the average is -0.49 standard deviations (2 basis points) and
is statistically signiﬁcant with a p-value less than 0.01. In other words, the FOMC announcement
leads to a resolution of uncertainty on average. But there is also a fair amount of variation across
individual FOMC dates, with some large declines and even some large increases.

Importantly, there is typically a direct relation between a speciﬁc change in the forward guidance
language used by the Federal Reserve and changes in our measure. To aid in interpretation, the
ﬁgure labels the three biggest falls and three biggest increases and provides a snippet from the
FOMC statement or related market coverage that helps understand these episodes. For example,
the biggest fall in our sample is in December 2008, where the FOMC cut rates to reach the zero
lower bound. But in addition to this rate cut, there was explicit forward guidance language (“...
warrant exceptionally low levels of the federal funds rate for some time") that signaled to the
market that low rates were here to stay. The second biggest fall is in August 2011. Prior to this
meeting, the FOMC statement contained a phrase that rates would be kept low “...for an extended
period". At the August meeting, the FOMC explicitly changed the language to signal that rates
would be kept low ”at least through mid-2013", the start of so-called “calendar-based forward
guidance". Markets clearly interpreted this as a sign that rates would indeed stay low and revised
downwards their uncertainty about future rates. Another date labeled on the ﬁgure is January 29,

76

2004. At this meeting there was a change in the language to “can be patient in removing its policy
accommodation" from the previous statement which said “accommodation can be maintained for
a considerable period". The market interpreted this as increasing uncertainty about when rates
would eventually increase.

In Section 3.4.2 below we discuss that changes in uncertainty are positively correlated with
ﬁrst-moment surprises. But the biggest changes in uncertainty do not always coincide with surprises
about the policy path, and vice versa. For example, among the four announcements with the largest
changes in uncertainty, two of them (in October 2008 and December 2008) also led to substantial
ﬁrst-moment surprises, whereas the other two (in August 2011 and November 1998) caused only
modest ones. Moreover, we ﬁnd that our measure of uncertainty (but not the ﬁrst-moment shock)
is positively correlated with changes in the dispersion of survey forecasts. In Table G12, we show
this using data from the Survey of Professional Forecasters (SPF) and the Summary of Economic
Projections (SEP).

Overall, our narrative evidence suggests that Fed communication has important eﬀects on
perceived monetary policy uncertainty and that these changes in uncertainty are often a separate
dimension of the Fed’s policy actions. We will systematically evaluate how these changes in
market-perceived uncertainty about the future rate caused by FOMC announcements is transmitted
to international ﬁnancial markets and document that they indeed constitute an important additional
dimension of FOMC actions.

3.2.2 Monetary Policy Surprise

As mentioned above, we also consider the well-known measure of a ﬁrst moment monetary policy
shock. This is because surprises to the expected path of policy rates (i.e. mps) are positively
correlated with changes in uncertainty about future rates (mpu), as can be seen in Figure H4. Thus
to isolate the eﬀect of monetary transmission through changes in the second moment we need to
control for changes in the ﬁrst moment. We do that in our analysis by using the following ﬁrst
moment measure, labeled mps or MP Surprise. This shock is calculated as the change in the futures

77

t = p(h)

t − p(h)

t

t−ε where p(h)

price in a window around the FOMC meeting. Let et represent the monetary shock and ε represent
the length of the window, then e(h)
is the price of a futures contract
at time t that matures in t + h. As with mpu our baseline measure uses a two-day window. We
use four Eurodollar futures contracts, expiring 1 quarter ahead (ED1) to 4 quarters ahead (ED4).13
Taken together, the four contracts contain rich information about the short and medium term path
of expected interest rates. To summarize this information in a parsimonious way we perform a
principal component analysis. The ﬁrst principal component of the 4 futures price changes explains
more than 90% of the total variation across all the contracts. We therefore use this ﬁrst principal
component as one of our measures of monetary policy shocks. This is essentially identical to the
measure used in Nakamura & Steinsson (2018). Since the scale of this principal component is
arbitrary, we normalize our measure to have a one standard deviation eﬀect on the 1 year ahead
rate.

One issue worth noting for both measures of monetary policy shocks is that the underlying
interest rate for Eurodollar futures is the three month LIBOR rate. LIBOR typically trades at a
spread over the federal funds rate; thus, our monetary shock measures capture not only changes in
the ﬁrst and second moments of the future policy rate but also changes owing to the time-varying
spread. The diﬀerence between LIBOR and the fed funds rate is best measured by the LIBOR-OIS
spread. Other than the period around the ﬁnancial crisis of 2007-2009, this spread has been low,
and crucially, stable. Moreover, as we discuss in the section on robustness checks below (Section
3.4.6) the results are unchanged when we control for this spread in the regression analysis.

3.3 Data

Next, we describe the data used to measure the international spillover of US monetary policy.
Our primary outcome measures are international asset prices: 2 year and 10 year government bond
yields, equity prices and exchange rates vis-à-vis the US. We collected this data from Bloomberg
for 28 advanced countries and 16 emerging market countries between January 1995 and June 2019,

13In Section 3.4.6 we show that our results are robust to using longer-horizon measures of mps.

78

where available. The data availability varies by country.14 Table G13 details the data coverage
for our international asset price data and the classiﬁcation of countries. We will focus on two day
changes around FOMC announcements. We use two day changes to allow for all international
asset markets to respond to US monetary policy shocks.15 The two day change is calculated as the
diﬀerence between the closing price one day after and one day before an FOMC announcement.
Some markets are open when FOMC announcements are made while others are not. We account
for these country-speciﬁc timing diﬀerences when calculating the two day changes. Below, we also
show that our results are robust to using a narrower one day window.

For constructing both of our US monetary policy shocks, we use daily Eurodollar futures data
and daily Eurodollar options data which are from the Chicago Mercantile Exchange. For US data,
the treasury yields are zero-coupon yields from Bloomberg and the S&P 500 return is from Yahoo
Finance.

3.3.1 Summary Statistics

Table G1 reports summary statistics for the key variables of interest. Panel (a) contains our
measures of monetary policy surprise and monetary policy uncertainty. The full sample includes
204 FOMC announcements from January 1995 to June 2019.16 To aid in interpreting the regressions
coeﬃcients, we normalize the two monetary policy shocks in the following way. The monetary
policy uncertainty measure is standardized to have unit standard deviation. Since the monetary
policy surprise measure is calculated using a principal component analysis, its scale is arbitrary;
thus, we normalize it to have an eﬀect on the 1 year ahead futures rate equal to one standard
deviation. Panel (b) presents summary statistics for exchange rate return, stock return, and changes

14We have data on exchange rate and stock prices for 28 advanced countries and 16 emerging countries. For government

yields we have data for 22 advanced countries and 8 emerging countries.

15This is consistent with the recent literature that uses two-day changes (Albagli et al. (2019) and Hanson & Stein
(2015b)) and is based on the idea that since FOMC meetings happen at 2:15 pm, using daily changes does not give
markets enough time to react before close.

16We exclude the 9/17/2001 announcement following the 9/11 terrorist attacks and the 5/2/2018 announcement due to a
lack of asset data availability. The 10/8/2008 announcement is excluded because many other central banks took joint
action on that date, making it impossible to isolate the eﬀect of US monetary policy. The 5/22/2013 “taper tantrum"
episode is excluded as well, as it was driven by a speech by the Chairman rather than an FOMC announcement.

79

in 2 and 10 year government bond yields for the advanced and emerging countries in our sample,
calculated in a two day window around FOMC announcements.

3.4 Results

In this section we present the main results that show the spillover eﬀects from changes in
US monetary policy uncertainty to international asset prices. To establish a benchmark, we ﬁrst
document the response of US asset prices to monetary policy uncertainty changes.
In Section
3.4.2 we present our main results for international bond yields, followed by an investigation of
the mechanism behind our baseline results in Section 3.4.3. Next, we provide a discussion of the
heterogeneity in the country-level responses in Section 3.4.4. This is followed by results on the
response of international equity markets in Section 3.4.5 and we conclude with robustness checks
in Section 3.4.6.

3.4.1 Response of US asset prices to US monetary policy uncertainty

We study the response of 2 and 10 year Treasury bond yields and the S&P 500 return. The
two monetary policy shocks detailed in Section 3.2 are i) monetary policy surprise (mpst) which
measures surprise changes in the expected path of the policy rate and ii) monetary policy uncertainty
(mput) which measures surprise changes in the uncertainty around the expected path, both for the
1 year horizon. As mentioned above, both measures have been scaled to reﬂect a one standard
deviation eﬀect. For each asset, we calculate the change in a two-day window, labeled (yt). We
estimate the following regression equation:

yt = α0 + α1mpst + α2mput + εt

(3.1)

The results are presented in Table G2 with heteroskedasticity-robust standard errors in parentheses.
The top panel shows the results for the full sample that runs from January 1995 to June 2019.
The middle panel shows a pre-crisis sample from January 1995 to November 2007 and the bottom
panel shows the post-crisis sample from December 2007 to June 2019. The ﬁrst row shows the
well-known eﬀect of monetary policy surprises on US ﬁnancial markets. A contractionary surprise

80

lowers stock prices and raises both 2 and 10 year yields.17 Results for the two samples are quite
similar with somewhat smaller eﬀects in the post-crisis sample.

The second row shows the response to monetary policy uncertainty shocks. An increase in
monetary policy uncertainty lowers stock prices and raises long-term bond yields, but not short-
term yields. For the full sample, the response of the 10 year yield is signiﬁcant (at the 1% level)
but the stock market and 2 year yield responses are not. Qualitatively, the eﬀects of an increase in
uncertainty have similar eﬀects to a contractionary monetary policy surprise. In terms of magnitude,
the response to a one standard deviation increase in mpu is about half the size of mps for stocks and
10 year yields. For example the 10 year yield increases by 4 basis points (0.3 standard deviations)
in response to a one standard deviation increase in mpu. Finally, the eﬀect of mpu strengthens in
the post-crisis sample. In the pre-crisis sample mpu only has a statistically signiﬁcant eﬀect on 10
year yields. But in the post-crisis sample, stock returns and 10 year yields respond signiﬁcantly to
mpu, with a rise in the size of the eﬀect as well. These results of the additional eﬀect of mpu on
US ﬁnancial markets are consistent with those documented in Bauer et al. (2019). We now turn
our attention to the main focus of this paper: the spillover eﬀects of mpu to international ﬁnancial
markets.

3.4.2 Response of international bond yields to US monetary policy uncertainty

The common event-study approach in the literature involves regressing an asset price on the
monetary policy shock measure in the event window. For studying the international spillover this
would translate to the following panel regression

yi,t = δ0 + δ1mpst + νi,t

(3.2)

where yi,t is the two-day change in asset price of country i on date t, with the monetary shock
measured by the so-called monetary policy surprise measure (mps) in a window around the FOMC
announcement. mps typically measures surprise changes in the expected path of the policy rate.

17The response of the stock price to mps is smaller and less signiﬁcant than Bernanke & Kuttner (2005). This result

in the more recent sample is due to our use of daily futures data as also noted by Lakdawala & Schaﬀer (2019b).

81

The main goal of this paper is to evaluate the response to changes in monetary policy uncertainty.
To do this we augment the above speciﬁcation by adding mpu, which measures surprise changes in
the uncertainty around the expected path. The regression takes the following speciﬁcation18

yi,t = β0 + β1mpst + β2mput + εi,t

(3.3)

where yi,t is the two-day change in the 2 or 10 year bond yield around the FOMC meeting on day
t for country i. As discussed in Section 3.2, both monetary shock measures are calculated for a 1
year horizon and have been scaled to reﬂect a one standard deviation eﬀect. We also standardize the
international bond yields to have unit standard deviation within each country.19 Since our variables
are measured within a relatively tight window around FOMC days, we make the assumption that
the FOMC announcement is the primary driver of asset prices in this window. This assumption
is commonly used in the event-study literature. While our baseline window is a two-day window
following Hanson & Stein (2015b) and Albagli et al. (2019), we show in Section 3.4.6 below that
our results are robust to using a one-day window.

For our main speciﬁcation we separate the countries into two groups: advanced and emerging
market countries.20 Table G3 reports the regression coeﬃcients with the advanced country results
in the top panel and emerging country results in the bottom panel. Standard errors reported
in parentheses are two-way clustered along the country and time dimension. In column (1) we
estimate Equation 3.2 with only mps as the regressor to document that monetary policy surprises
have a statistically signiﬁcant and economically meaningful impact on international yields in both
advanced and emerging countries. A contractionary monetary policy suprise in the US raises bond
yields around the world. This result is well established in the literature for both advanced and
emerging countries, e.g. see Hausman & Wongswan (2011), Gilchrist et al. (2019) and Albagli
et al. (2019).

18As we discuss in the section on robustness checks (Section 3.4.6), adding country ﬁxed eﬀects to this baseline

19This is done because there is substantial heterogeneity in the standard deviation of yield changes across countries in

speciﬁcation does not change our results.

our sample. This can be seen in Figure H5.

20 We have yield data for 22 advanced and 8 emerging countries. See Table G13 for details on the sample countries.

82

The second column in each panel adds mpu to the regression and shows our ﬁrst set of
main results. For both advanced and emerging countries, the mpu shock has a statistically and
economically signiﬁcant eﬀect on 2 and 10 year bond yields. An increase in monetary policy
uncertainty raises global bond yields even after controlling for the conventionally used ﬁrst-moment
shock (i.e. mps). This eﬀect is bigger for 10 year yields compared to 2 year yields and is also bigger
for advanced countries relative to emerging countries. A one standard deviation increase in mpu
raises 10 year yields in advanced (emerging) countries by .264 (.171) standard deviations.21 At the
long end of the yield curve, mpu has a bigger eﬀect than mps, while at the shorter end the mps
eﬀect is larger. Additionally, for advanced countries, an increase in monetary policy uncertainty
raises 10 year yields by roughly twice as much as 2 year yields, but for emerging countries the
response of 2 and 10 year yields is essentially the same.

While the average eﬀect of mpu on international bond yields is moderate, we show that this can
amount to a substantially larger role for mpu in driving bond yields on days with big changes in mpu.
Moreover, the eﬀect of mpu dwarfs the mps eﬀect on these days. As we discussed in Section 3.2.1,
the big changes in mpu occur when the Federal Reserve willfully chose to make notable changes to
the forward guidance language used in the FOMC statement. To study the yield response on these
“prominent" dates, we isolate the ten FOMC dates with the largest increase in mpu and the ten
FOMC dates with the largest decrease in mpu. We then average the total change in 10 year bond
yields on these dates and compare to the average predicted component due to mpu and that due
to mps, based on the coeﬃcients estimated in Equation 3.3.22 Figure H2 plots this separately for
advanced and emerging countries. For advanced countries, the white bar shows that yields fall (or
rise) by about one standard deviation (8 basis points) on these dates. mpu (gray bar) accounts for
nearly half of this change on days with large increases in mpu and nearly all of the change on days
with large decreases. In contrast, mps (black bar) accounts for less than one quarter. For emerging
countries, the pattern is similar: mpu accounts for a lion’s share of the change in bond yields on
these 20 dates. From the 15 basis point fall in emerging yields, roughly three-quarters is due to

21This amounts to a 2 basis point rise in advanced and 3 basis point rise in emerging countries.
22In Figure H6 we show numbers for each of the 20 FOMC meetings individually.

83

mpu. Thus, Federal Reserve actions that aﬀect the market’s perceived uncertainty about future
rates can have substantial eﬀects on global yields. The implication is that the Federal Reserve has
an additional tool in its arsenal to aﬀect international ﬁnancial conditions, one which has scarcely
received any attention in the literature.

To contextualize the magnitude of the mpu spillover, we estimate the eﬀects of “news shocks"
around major US macroeconomic data releases.23 Table G4 shows that retail sales shocks have a
statistically signiﬁcant eﬀect on advanced country 2 and 10 year yields, and that CPI shocks have a
signiﬁcant impact on 2 year yields. Importantly, the yield response is much larger for the monetary
policy shocks, as a one standard deviation change in mpu has an eﬀect on international bond yields
that is at least one order of magnitude higher than the eﬀect of news shocks. Furthermore, the table
also shows that none of the news shocks have a statistically signiﬁcant eﬀect on emerging country
yields. These results highlight the unique impact of US monetary shocks on international yields.
We also check if the eﬀect of mpu on international asset prices is driven by speciﬁc announce-
ments about large scale asset purchases (quantitative easing or QE). Table G14 shows the baseline
regression from Equation 3.3 where we have added a QE dummy along with its interaction with
mps and mpu. The QE dates are taken from Fawley et al. (2013). The results show that yields in ad-
vanced countries do not respond diﬀerently to mpu on these dates and there is some weak evidence
that the eﬀects of mpu are stronger for 10 year yields in emerging countries. More importantly,
the baseline eﬀect of mpu on yields on non-QE dates stays roughly the same in terms of economic
size and statistical signiﬁcance. There is a debate in the literature on whether QE transmits through
the signaling channel (Bauer & Rudebusch (2014)). Our results suggest that QE can also transmit
through an “uncertainty signalling channel", whereby signals about uncertainty regarding future
rates have an eﬀect on ﬁnancial markets. In Section 3.4.6 we also show that excluding the zero
lower bound period does not materially aﬀect our results.

23We collect data on ﬁve news announcements in the US: employment, GDP, CPI, PPI, and retail sales. For each
news release, the surprise component, or “shock”, is calculated as the diﬀerence between the actual released number
and the consensus forecast from Action Economics/Money Market Services. For the employment report, we use
non-farm payrolls, for CPI and PPI we use headline inﬂation, retail sales are the total sales including automobiles
and GDP is the advance GDP release. We scale the news shocks to have unit standard deviation so that the size of
the coeﬃcient can be directly compared to the mpu coeﬃcient.

84

Next, in Section 3.4.3 below we investigate the mechanisms driving the international yield re-
sponse to monetary policy uncertainty, including diﬀerences across countries and across maturities.
But ﬁrst, we document that there have been some important changes over time in the transmission
of mpu to international asset prices, especially for the emerging countries.

Table G15 and Table G16 report estimates for the baseline speciﬁcation from Equation 3.3,
splitting the full sample into a pre-crisis sample that runs from January 1995 to November 2007 and
a post-crisis sample from December 2007 to June 2019. For our split sample results, we perform
the standardization of our variables based on the split sample standard deviations. For advanced
countries, the yield response is roughly similar across the samples, with a slightly larger response
of the 10 year yield in the post-crisis sample. For emerging countries, however, a diﬀerent picture
emerges. The signiﬁcant response of emerging yields to mpu in the full sample is driven entirely
by the post-crisis sample. Speciﬁcally, the yield response is insigniﬁcant and essentially zero in
the pre-crisis sample but larger in magnitude and strongly signiﬁcant in the post-crisis sample. We
discuss below in Section 3.4.4 that this can partially be explained by the average increase in capital
account openness in emerging countries over the full sample.

We also investigate the dynamic response of bond yields using a local projection framework
(Jordà (2005)). These results are presented in Figure H7. While 2 and 10 year yields remain
elevated in both advanced and emerging countries after the monetary policy uncertainty shock, the
standard errors get large after around a month. This makes sense as there is more background noise
(i.e. other events that drive yields) as the horizon gets longer. Thus, in this paper we focus on the
precisely estimated higher frequency impact.

Table G3 also shows that accounting for mpu is crucial in assessing the transmission of FOMC
actions to international bond yields, even if one is only interested in the conventional monetary
policy surprises (mps). Leaving out mpu biases both the coeﬃcient estimate of the yield response
to mps and the unconditional mean, as can be seen by comparing the mps coeﬃcient and the
intercept in columns (1) and (2). This can be easily understood from a basic omitted variable
analysis. Column (1) estimates the speciﬁcation from Equation 3.2 with mpu being the omitted

85

variable. The key point is that mps (i.e. surprises to the expected path of policy rates) is positively
correlated with mpu (i.e. changes in uncertainty about future rates). We document this correlation
in Figure H4. This correlation results in an upward bias because the sign of the correlation between
the dependent variable and mpu is the same as that between mps and mpu (both correlations are
positive, as can be seen from the table). The table shows that leaving out mpu overstates the mps
eﬀect by roughly 50%.24

In addition to biasing the coeﬃcient on mps, omitting mpu also aﬀects the estimate of the
intercept. Estimates of the intercept term from column (1) are negative and signiﬁcant for both
yields and both sets of countries. This systematic average decline is at odds with the assumptions
of the event-study framework where changes in asset prices around FOMC meeting should be
unpredictable. However, once we add mpu to the regression, the intercept term becomes eﬀectively
zero and statistically insigniﬁcant. This is driven by an average decline in mpu documented in
Figure H1. Thus leaving out mpu means that the average fall in mpu is soaked up in the intercept,
making it turn negative.25

Yet another way to see the importance of accounting for uncertainty changes is by comparing the
R2 in the two columns. There is a substantial increase in R2 with the addition of mpu, especially at
the longer end of the yield curve. For example the R2 for emerging country 10 year yields increases
from 0.036 to 0.06.

Thus our results suggest that the literature on the spillover eﬀects has likely overestimated
the eﬀect coming purely through monetary policy surprises (ﬁrst-moment shocks) while under-
estimating the total eﬀects of US monetary policy actions which also work substantially through
second-moment (or uncertainty) changes.

24Since mpu is a risk-neutral measure it captures both the quantity and price of uncertainty.

If mps shocks drive
risk-aversion (or risk compensation) in ﬁnancial markets including option prices and mpu, our methodology of
controlling for mpu would be under-estimating the true mps eﬀect. On the other hand it seems reasonable to expect
that uncertainty or mpu shocks (and not mps shocks) are more likely to aﬀect risk-aversion. Since estimating the
risk-aversion response is out of the scope of this paper, we caution the reader here and recommend viewing the 50%
estimate as an upper bound. Regardless, this issue is not crucial for interpreting the eﬀects of mpu since we always
control for mps when reporting estimates of mpu.
25To make this clear, consider decomposing mput into a constant and time-varying term, mput = µmpu + εt,mpu. If
the true model is Equation 3.3 but mpu is omitted from the regression, then the residual will soak up β2 ∗ εt,mpu
and the intercept will soak up the term β2 ∗ µmpu which is negative since β2 > 0 and µmpu < 0.

86

3.4.3 Understanding the response

What explains the response of international bond yields to changes in US monetary policy un-
certainty? In this section, we provide evidence that an international portfolio balance channel is
primarily responsible for transmission to advanced countries whereas a ﬂight to safety channel is
likely driving the emerging country response.

We start by discussing the mechanism through which mpu aﬀects bond yields domestically in
the US. There is a clear risk-based explanation for the transmission mechanism. Standard asset
pricing theory implies that expected excess returns depend on the negative covariance of returns
with the stochastic discount factor. One factor driving this covariance is the uncertainty about future
returns, as can be seen by rewriting the covariance in terms of the correlation and the standard
deviations:

Et Rt+1 − R f

t =

=

Covt(−Mt+1, Rt+1)
Corrt(−Mt+1, Rt+1)

Et Mt+1

Et Mt+1

σt(Mt+1)σt(Rt+1)

Since changes in mpu are a clear signal about the conditional volatility of bond returns (σt(Rt+1)),
then higher short-rate uncertainty should raise term/risk premia.In earlier work, Bauer et al. (2019)
and Bundick et al. (2017a) show that changes in mpu do indeed transmit to US bond yields through
changes in term premia. In Table G17 we document this eﬀect using the diﬀerent term premium
estimates of Joslin et al. (2011), Adrian et al. (2013) and Kim & Wright (2005). For all three
measures, the table shows that an increase in uncertainty raises term premia on US bond yields.26
Next, we study whether mpu aﬀects the term premia even in global yields. We apply the
methodology of Joslin et al. (2011) to carry out the decomposition into an expected component
and term premium component.27 Table G5 shows the results where we use the same speciﬁcation

26While our results about the role of uncertainty for term premia are evident, we cannot rule out that the uncertainty
about the stochastic discount factor or the correlation between future returns and the stochastic discount factor could
also be important.

27Since we have only zero-coupon yield data from Bloomberg, it is not straightforward for us to implement the

alternative procedures in the literature, for example the measure of Adrian et al. (2013).

87

from Equation 3.1 now separately using the expected component and term premium component as
the dependent variables.28

For advanced country yields, the response to mpu is entirely due to the response of the term
premium with no response of the expected component. However, for emerging country yields, this
pattern is reversed. The response to mpu is not driven through the term premium but rather through
changes in the expected component. In other words, in response to changes in monetary policy
uncertainty, the market does not expect that central banks of advanced countries will respond by
changing their own policy rates; however, they do expect that central banks of emerging countries
will. Moreover, markets react in a way that suggest their perception of the term premium for holding
advanced country bonds has changed but not for emerging country bonds.

Since we are studying the response of bonds denominated in local currencies, we might expect
movement in the exchange rate to be playing a role. Table G6 shows the response of the exchange
rate to mps and mpu for our full sample. The exchange rates are in units of foreign currency per
US dollar and thus an increase represents a depreciation of the foreign currency relative to the US
dollar. As documented in the literature a contractionary mps leads to an appreciation of the U.S
dollar relative to both advanced and emerging countries (e.g. see Hausman & Wongswan (2011)).
However, we again note that including mpu in the regression lowers the estimated eﬀect of mps.
More importantly, we notice that an increase in mpu depreciates currencies of emerging countries
but not those of advanced countries. The advanced country response is consistent with the results
of Gilchrist et al. (2019) who ﬁnd a substantial eﬀect of US monetary policy on dollar-denominated
sovereign bonds. They label this channel the “ﬁnancial spillover" channel. In the next section,
we provide a more detailed discussion and evidence for understanding the speciﬁcs of the mpu
spillover to advanced countries. For emerging countries, we document below that this result is
consistent with the recent framework of Rey (2013), where even exchange rates being “ﬂexible"
does not insulate the country from the ﬁnancial spillover. However, we show countries with more
capital account restrictions respond less. This discussion is presented in Section 3.4.3.2.

28Note that while mps and a constant are included in the regression, the coeﬃcients are left out for space considerations.

88

3.4.3.1 Response in advanced countries

Why does the bond term premium of advanced countries (but not emerging ones) respond to mpu?
We provide evidence for an international portfolio balance channel whereby mpu transmits through
changes in the term premium on US bonds to term premia of foreign country bonds; but only for
the bonds of countries that are considered substitutes for US bonds.

We ﬁrst show that mpu transmission to term premia in advanced countries is driven primarily
through the eﬀect of mpu on US bond term premia. To do this we take the speciﬁcation reported
in Panel (a) of Table G5 (i.e. our baseline regression from Equation 3.3 but with the term premium
as the dependent variable) and control for the change in the term premium on the US 10 year bond.
Then we compare the coeﬃcient on mpu from this speciﬁcation (reported in Panel (b) of Table
G5) to one without the US 10 year term premium (reported in Panel (a)). If there is no change
in the mpu coeﬃcient from Panel (a) to (b), then we can conclude that none of the mpu eﬀect is
working through US term premium changes. On the other hand, if the coeﬃcient goes to zero,
then we can conclude that the mpu eﬀect is working through US term premium changes. The
table shows that controlling for the US term premium has a signiﬁcant eﬀect on the term premium
response to mpu, but only for advanced countries. Speciﬁcally, the mpu coeﬃcients that govern
the term premium response of emerging countries are essentially unchanged: they remain close to
zero and statistically insigniﬁcant. But for advanced countries, the mpu coeﬃcient of the 10 year
term premium response, which was substantial (0.24) and strongly signiﬁcant, drops to close to
zero (0.07) and is not statistically signiﬁcant. The same pattern holds for the 2 year term premium
response. This is suggestive evidence that changes in monetary policy uncertainty that drive term
premia in US bonds also transmit to term premia in international bonds as well, although only in
advanced countries.

Why do term premia in advanced countries respond to mpu-driven changes in the US term
premium but term premia of emerging countries do not? There is not much theoretical work in
the literature on the spillover eﬀects through US monetary policy uncertainty.29 The theoretical

29There is some recent work on the spillover of overall US uncertainty, see for example Bhattarai et al. (2019). But

89

paper closest to our empirical work that explicitly studies this topic is the recent work of Ghironi &
Ozhan (2019). However, in their paper they study only the spillover of monetary policy uncertainty
to emerging countries and allow the international trade of short-term securities only. Thus there is
no clear implication from their model for the diﬀerential response of long-term yields in emerging
versus advanced countries. However there is a much larger literature that studies the spillover of
US unconventional monetary policy, see Bhattarai & Neely (2016) for a recent survey. A common
theme in this literature is that US unconventional monetary policy aﬀects the term premium on US
bond yields and also has eﬀects on international bond yields. More speciﬁcally, the recent DSGE
model of Alpanda & Kabaca (2019) that features an international portofolio balance channel is
especially relevant for understanding our empirical results. They take the portfolio balance channel
that features imperfect substitutability between bonds of diﬀerent maturities and extend it to have
imperfect substitutability between domestic and foreign bonds. A prediction from this framework
is that term premium changes in the US should aﬀect term premia in foreign countries based on
how substitutable the foreign country’s bonds are with the US. Speciﬁcally, a higher elasticity of
substitution for a given country will mean a larger response of that country’s term premium. We
now provide evidence that this is indeed the channel through which mpu has international spillover
eﬀects.

If the long-term bonds of two countries are highly substitutable, this implies the bonds share
similar degrees of risk and generally belong to the same class of securities. Perceived risk and
the demand for a class of securities are two main determinants of a bond’s term premium; thus,
the term premium of two highly substitutable bonds should positively comove.30 Accordingly, we
construct a simple measure of the substitutability of bonds by calculating the correlation between
the 10 year bond term premium of each country in our sample and the US on all non-FOMC days
between January 1995 and June 2019. Our measure is essentially identical to the method of de los
Rios & Shamloo (2017) and is comparable to the empirical test of Frankel (1982). After calculating

this work does not try to isolate the eﬀect of monetary policy uncertainty.

30The models of Kabaca (2016) and Alpanda & Kabaca (2019) generate the feature that high substitutability between
domestic and foreign bonds results in strong positive comovement between domestic and foreign term premia. See
Bernanke (2015) for further discussion of the term premium and the factors that move it.

90

the correlation between the 10 year term premium for country i and the US, we scale our measure
to lie between 0 and 1, i.e. we add 1 to each correlation and divide by 2.31 Our speciﬁcation is as
follows

yi,t = κ0 + κ1mpst + κ2mput + κ3mpst ∗ bondsubi + κ4mput ∗ bondsubi + εi,t

(3.4)

Since the substitutability measure is scaled to lie between 0 and 1, the coeﬃcient on the mpu
interaction term can be interpreted as the marginal eﬀect of an mpu shock on the 10 year term
premium of a country with perfect substitutability (i.e. correlation= 1), relative to a country with
the least substitutability (correlation= −1). Table G7 displays the mpu coeﬃcients (κ2 and κ4) for
all countries pooled together, only advanced countries and only emerging countries. For the pooled
sample, the coeﬃcient on the interaction term is positive and signiﬁcant. This implies that it is
indeed the case that countries whose bonds are more substitutable with US bonds on non-FOMC
days also display a larger term premium response to changes in US monetary policy uncertainty
on FOMC days. We see the same pattern when we restrict the regression to the advanced country
sample but not when we do that for emerging countries. Advanced countries whose bonds are
more substitutable with US bonds are also more sensitive to mpu shocks. We have also plotted the
distribution of our bond substitutability measures. As expected, the bonds of advanced countries
have noticeably higher substitutability with US bonds than do emerging market countries. The
ﬁgure is omitted for space considerations.

To summarize, the response of advanced country bond yields to mpu is driven by an international
portfolio balance channel with the high degree of substitutability between advanced country bonds
with the US being the crucial factor. Next, we turn to why emerging country bonds also respond
signiﬁcantly to mpu.

31For robustness, we also calculate the bond substitutability measure using non-FOMC days between January 1995
and the FOMC meeting on day t. Online Table G25 shows the results for this alterrnative bond substitutability
measure. Our results are also robust to using changes in the 10 year term premium (rather than levels) to compute
the correlations and alternatively using a logistic transformation or by using a non-linear speciﬁcation, e.g. binning
by quantiles.

91

3.4.3.2 Response in emerging countries

In the previous section we discussed the international portfolio balance channel to understand the
response of advanced country yields to mpu. But what explains the transmission of mpu through
the expected component of bond yields for emerging markets? In this section we provide evidence
on the response of capital ﬂows in emerging countries that is consistent with a ﬂight to safety
channel.

We use data from the Treasury International Capital (TIC) reporting system, following the
methodologies of Bertaut & Tryon (2007) and Bertaut & Judson (2014). Bertaut & Tryon (2007)
construct a monthly measure of US holdings of foreign securities by combining annual TIC survey
data with monthly TIC S ﬂow data. Bertaut & Judson (2014) improve upon the measure by
incorporating monthly TIC SLT holdings data, which becomes available in December 2011. We
use the Bertaut & Tryon (2007) measure from 1995 to 2011 and the Bertaut & Judson (2014)
measure from 2012-2018.32

This data only measures holdings by US residents of foreign assets; thus, our analysis is limited
and we cannot observe the response of non-US investors to the monetary shocks. Another important
caveat is that the TIC data are available at a monthly frequency and thus our regression speciﬁcation
is not as clean as the higher frequency speciﬁcation used in the rest of the analysis. Nevertheless,
we believe the data provides some interesting evidence.

We study the response of US holdings of foreign bonds to mps and mpu together with the
interaction of the diﬀerence in the 3 month interest rates of the foreign country and the US (labeled
idi f f )

yi,t = γ0 + γ1mpst + γ2mput + γ3mpst ∗ idi f fi,t + γ4mput ∗ idi f fi,t + εi,t

(3.5)

The sample runs from 1995 to 2018 for a total of 187 FOMC meetings and excludes the ﬁnancial
crisis period from December 2007 to June 2009.33

32The data can be accessed here: https://www.federalreserve.gov/econres/ifdp/2014.htm. See Bertaut & Tryon (2007)

and Bertaut & Judson (2014) for details.

33The literature has documented a large and abnormal reduction in international capital ﬂows during the Great
Recession, and in crisis periods generally. See, for example, Milesi-Ferretti & Tille (2011) and Broner et al. (2013).

92

Table G8 shows that for emerging countries an increase in mpu leads to a reduction in bond
holdings. Moreover, the interaction coeﬃcient between mpu and the interest rate diﬀerential is
negative and signiﬁcant. This means that more capital ﬂows out of emerging countries which have
a higher interest rate diﬀerential with the US (consistent with the results in Ahmed & Zlate (2014)).
Table G26 conﬁrms this result holds when including time ﬁxed eﬀects in the speciﬁcation. Overall,
these results are consistent with a ﬂight to safety/quality channel where investors are pulling money
out of countries that are perceived to be riskier than the US. These results are consistent with
Bhattarai et al. (2019) who use a VAR framework and also ﬁnd a ﬂight to safety response of
emerging countries to an increase in US uncertainty. They use VIX to capture a broader measure
of US uncertainty. However as discussed in Bauer et al. (2019), mpu drives a substantial amount
of variation in the VIX on FOMC meeting days, which is the sample that we focus on here. Thus
our results are pointing to the role of US monetary policy speciﬁc uncertainty in driving this result.
In addition to the mechanism described above, there is a testable implication of the emerging
country expected component response to changes in mpu. In response to an increase in monetary
policy uncertainty, ﬁnancial markets are expecting interest rates to rise in the future in emerging
countries but not in advanced countries. Assuming that the markets are not systematically wrong,
we can check to see if short rates do indeed move in the expected direction. We test this implication
using the local projections framework outlined in Section 3.4.2 to map out the dynamic response
of 3 month bond yields.

Results from this exercise are presented in Figure H3 with conﬁdence intervals that use Driscoll-
Kraay standard errors. Short rates in emerging markets do indeed move in the direction markets
expect, as the 3 month yield is signiﬁcantly higher for most of the time in the one and a half years
after the mpu shock.34 In contrast, the 3 month yield shows essentially no change a year after the
mpu shock in advanced countries. Thus, it appears markets are correctly expecting short rates to
respond to US monetary policy uncertainty in emerging countries, but not in advanced. These
results reinforce that uncertainty is transmitted to advanced and emerging bond yields through

34For example, after a year short rates are 0.15 standard deviations or 3.8 basis points higher.

93

fundamentally diﬀerent mechanisms.

We also ﬁnd that the emerging market response to mpu is suggestive of the framework put forth
by Rey (2013). As shown above in Table G6, exchange rates of emerging countries depreciate in
response to a contractionary mpu shock. From a Mundell-Fleming perspective, a ﬂexible exchange
rate should be enough to shield a country to ﬁnancial spillovers from the US. Rey (2013) suggests
that this would only be the case if there are additional restrictions on capital mobility. In the next
section we show that it is indeed the case that emerging countries whose capital account is more
unrestricted are the ones that respond more to mpu.

3.4.4 Heterogeneity in response to monetary policy uncertainty

In this section, we explore potential country characteristics that are associated with the sensitivity to
US monetary policy uncertainty transmission. Speciﬁcally, we test for diﬀerences in the response
of asset prices conditional on country characteristics.

To establish noticeable heterogeneity in asset price responses to mpu shocks, we ﬁrst plot the
country-speciﬁc estimated coeﬃcients on mpu for the full sample period. Figure H8 displays the
results for 2 and 10 year bond yields. Across both advanced and emerging countries, we see a
mix of statistically signiﬁcant and statistically insigniﬁcant responses. Within the two country
groups, the most positive responses are signiﬁcantly diﬀerent than the least positive responses.
This points toward a meaningful amount of heterogeneity that can potentially be explained, beyond
the advanced versus emerging distinction.

We attempt to explain this heterogeneity by using time-varying country characteristics. Using
time-varying, rather than ﬁxed, country characteristics allows us to use both within-country and
between-country variation in our identiﬁcation. Our baseline observables include ﬁnancial depth,
exchange rate regime, trade openness and the change in the 3 month interest rate diﬀerential
with the US on an FOMC day. For emerging market countries, we also include capital account
openness.35 Financial depth is the value of credit provided to the private sector, as a percentage

35The capital accounts of advanced countries are almost exclusively the maximum degree of openness. Thus, virtually

94

yi,t = α + β1mpst + β2mput +

i,tmpst)⊥ +

v∈V

1(ev
γv

v∈V

2(ev
γv

i,tmput)⊥ + i,t

(3.6)

of GDP. Exchange rate regime is deﬁned as in Ilzetzki et al. (2019), where the categories are a
ﬂexible exchange rate, a partial peg and a ﬁxed regime. Trade openness is the sum of total exports
and imports divided by GDP. Capital account openness is deﬁned as in Chinn & Ito (2006), with a
higher value indicating greater openness (i.e. fewer capital controls).

Using this set of observables, our methodology most closely follows that of Iacoviello & Navarro
(2019) by recursively orthogonalizing the regressors. Let v ∈ V be our set of time-varying country
characteristics for either the advanced country sample or the emerging market country sample. We
estimate the following equation:

where ev

i,tmpst)⊥ and (ev

i,t is the annual exposure index of variable v for country i on FOMC day t. The
i,tmput)⊥ are such that the β1 and β2 coeﬃcients measure
interaction terms (ev
the response to an mps and mpu shock, respectively, when the exposure indices are at their 25th
percentile values. The γv
2 coeﬃcients capture the marginal response to an mps and mpu
shock, respectively, when the exposure index ev

i,t is at its 75th percentile value.36

1 and γv

Following Iacoviello & Navarro (2019), the orthogonalized interaction terms are constructed in
the following manner. First, each exposure variable, v, is standardized, i.e. we subtract the sample
mean and divide by the sample standard deviation, to make the scale of the exposure indices more
comparable. Second, we perform a logistic transformation of the standardized exposure variables
to collapse the variables to a unit interval. Third, we scale the transformed exposure variables to
the distance between the 25th and 75th percentiles i.e. we subtract the 25th percentile value and
divide by the diﬀerence between the 75th and 25th percentile values. At this stage, we now have our
exposure indices, ev
i,t. Next, we multiply our exposure indices by each of the shocks to calculate our
interaction terms: (ev
i,tmpst) and (ev
i,tmput). For the ﬁnal step, we recursively orthogonalize each of
the interaction terms, starting with the mps interaction within each interaction term pairing. In other
no variation exists to explain the heterogeneity in response and we exclude capital account openness as an observable
variables for advanced countries.

36The 25th and 75th percentile values are calculated for the pooled (i.e. combined advanced and emerging) sample.

For exchange rate regime, this is equivalent to moving from a ﬂoating regime to a ﬁxed regime.

95

words, the interaction of the ﬁrst exposure variable (v1) and mps are orthogonalized with respect to
mpst and mput, while the interaction of v1 and mpu are orthogonalized with respect to mpst, mput
and the interaction of v1 with mpst. Then, (ev2
i,t mpst) is orthogonalized with respect to mps, mpu
and both of the orthogonalized v1 interaction terms, while (ev2
i,t mput) is orthogonalized with respect
i,t mpst)⊥. This continues for
to mps, mpu, both of the orthogonalized v1 interaction terms and (ev2
all subsequent exposure variables. In the following tables, the variables are orthogonalized with
respect to those that appear above them, e.g.
foreign exchange regime is orthogonalized with
respect to ﬁnancial depth and capital account openness in Table G9.

The above procedure has at least two advantages. First, the orthogonalization addresses the
within-country correlation between the diﬀerent characteristics. Without orthogonalizing, this
collinearity would impact the precision of our estimates.37 Second, since the orthogonalization
is recursive, each additional characteristic’s coeﬃcient can be clearly interpreted as a marginal
eﬀect after controlling for the previous characteristics. In theory, our choice of variable ordering
could aﬀect the results. We show below that our main results are robust to the ordering of
orthogonalization.

Focusing on the yield responses for emerging market countries, Table G9 shows that a more
open capital account signiﬁcantly explains diﬀerences in both 2 year and 10 year yield responses.38
Speciﬁcally, a country at the 75th percentile of capital account openness experiences an additional
0.12 standard deviation (0.088 standard deviation) increase in the 2 year yield (10 year yield) in
response to a US monetary policy uncertainty shock, relative to a country at the 25th percentile
value of capital account openness. The relationship between emerging country capital account
openness and responsiveness to US monetary policy uncertainty is consistent with the results of
Bowman et al. (2015) for US monetary policy surprises.39 As in Bowman et al. (2015), countries
with a greater degree of capital account openness are more sensitive to US monetary policy changes.

37Note, however, that our results are robust to orthogonalization.
38Table G18 contains the results for advanced countries. It shows that the change in the 3 month interest rate diﬀerential
with the US on an FOMC day is the only observable that signiﬁcantly explains heterogeneity (2 year yield response
only).

39Bowman et al. (2015) use monthly changes in 10 year US sovereign yields to identify monetary policy surprises and

monthly changes in emerging 10 year sovereign yields as the response variable of interest.

96

Other than capital account openness, a country’s exposure to dollar-denominated debt also
appears to matter for yield responses (see Table G19).40 The 10 year yield for countries with a
larger share of debt denominated in US dollars responds more sensitively to mpu shocks. Since an
increase in US monetary policy uncertainty is expected to appreciate the dollar, this increases the
real value of dollar-denominated debt and, thus, the likelihood of binding borrowing constraints.

Recursively orthogonalizing our variables allows us to estimate the marginal contribution of
each exposure variable, after we have controlled for any exposure variables that enter the regression
ﬁrst. As a result, the ordering of the variables can theoretically impact the results. In practice,
our main results are generally unaﬀected by the choice of variable ordering. The signiﬁcance
of capital account openness for the 2 year yield results does not depend on the ordering of the
orthogonalization. For the 10 year yield results, ﬁnancial depth must enter the regression prior to
capital account openness, but otherwise the ordering of the variables does not matter.41

3.4.5 Response of international equity indices to US monetary policy uncertainty

In this section we investigate the eﬀects of mpu on international stock markets. We use the same
speciﬁcation from Equation 3.3 with the 2-day return in the international equity indices as the
dependent variable. Table G10 shows the result for the full sample, pre-crisis sample ending in
November 2007 and a post-crisis sample starting in December 2007.

An increase in mpu leads to a fall in stock prices in both advanced and emerging countries. For
emerging countries the pattern is similar to the bond yield response: no eﬀect in the pre-crisis sample
but a strong and signiﬁcant eﬀect in the post-crisis sample. A one standard deviation increase in
mpu reduces stock prices by 0.65%. The advanced country response also follows this sub-sample
pattern, with essentially zero eﬀect in the pre-crisis sample and a 0.37% fall in the post-crisis

40Dollar debt exposure is the value of dollar-denominated debt as a % of GDP. Dollar debt exposure data is available
only through 2012 and the measure is signiﬁcant for 10 year yields only; thus, we did not include dollar debt exposure
in our baseline table.

41Figure H9 displays the coeﬃcient for the capital account openness interaction with mpu for all 24 unique variable
orderings with ﬁnancial depth listed ﬁrst. Note that the magnitude and statistical signiﬁcance of capital account
openness generally become stronger as capital account openness is orthogonalized with respect to more variables,
i.e. enters the regression later.

97

sample.42 Thus, the spillover of US monetary policy uncertainty to international equity markets
appears to largely be a post-crisis phenomenon. This result generalizes the pattern documented
in Lakdawala (2018) for Indian stock markets. We also tried to explain the heterogeneity in the
country-level response of equity markets using the country characteristics described above but did
not ﬁnd anything signiﬁcant. A potential channel could be the growing role of foreign institutional
investors in domestic equity markets as discussed in Lakdawala (2018) for Indian stock markets.
We leave this topic for future research.

As with the bond yield results, we see a similar pattern in the eﬀect of mps after accounting
for mpu. Focusing on the post-crisis sample, after including mpu in the regression, the response of
stock prices to mpu falls by about one-third in advanced countries and about one-half in emerging
countries. Moreover, accounting for mpu in the regression leads to roughly a doubling of the R2
for both advanced and emerging countries. This again highlights the importance of mpu even if
one is only interested in the transmission of US monetary policy through ﬁrst-moment shocks.

3.4.6 Robustness Checks

We conduct a variety of robustness checks for the results presented above. First, we show that our
results are not driven by the zero lower bound (ZLB) period from December 2008 to December
2015. Table G11 shows that results from the non-ZLB sample are very similar to the full sample
results. Next, we re-estimate equation 3.3 with the asset price changes and monetary shock measures
calculated over a one day window, rather than the two day window used in the baseline results.
Estimates using this narrower window are presented in Table G20 and show that overall the results
are essentially unchanged.

Since the construction of our measure of monetary policy uncertainty relies on Eurodollar
futures where the underlying interest rate is the LIBOR rate and not the Fed’s main policy tool
(federal funds rate), we want to make sure that instability in this spread is not driving our results.
The best way to measure this spread is using the LIBOR-OIS spread. To this end, we re-estimate

42Recall that advanced country bond yields responded to mpu both in pre- and post-crisis samples.

98

our baseline estimates from Equation 3.3 but also control for changes in this LIBOR-OIS spread.
The sample begins in December 2001 when LIBOR-OIS data is available. The estimates in Table
G21 show that our results are robust to this particular concern.

In this paper we have highlighted that, in addition to ﬁrst-moment eﬀects of mps, there is a
role of mpu through which US monetary policy can aﬀect international markets. In Tables G22
and G23 we show that our results are robust to using longer-term measures of mps: changes in
the 2 and 10 year Treasury yields, respectively. In earlier work, Gürkaynak et al. (2005) show that
two separate factors better characterize the ﬁrst-moment shocks (a target and a path factor), while
we just use the ﬁrst principal component (i.e. only one factor) to construct our mps measure. In
Online Table G27 we show that controlling for these two factors does not change the eﬀect of mpu
on international asset prices.

One concern with the FOMC day event-study approach is the issue of unscheduled FOMC
meetings. These are meetings outside the regular FOMC calendar and are typically responses to
unusual circumstances in the economy. In Online Table G28 we show that our results are robust to
excluding these unscheduled meetings. Another common concern is whether the results are aﬀected
by the so-called “information eﬀect" where the Federal Reserve signals its private information about
the underlying economic fundamentals. In Online Table G29 we control for this by cleansing the
monetary shocks using the methodology used by Campbell et al. (2012) and Lakdawala & Schaﬀer
(2019b). The results show that information eﬀects are not playing a role in driving the transmission
of mpu to international bond yields.

In the baseline speciﬁcation we do not include country ﬁxed eﬀects. In Online Table G30 we
present results including it and ﬁnd that the results are essentially unchanged. Finally, we explore
the sensitivity of our results to outliers across both countries and FOMC dates. Figures H10 and
H11 plot coeﬃcients and conﬁdence intervals from the baseline speciﬁcation while removing one
country and one FOMC date at a time, respectively. The coeﬃcients and conﬁdence intervals
remain similar regardless of which countries or dates are removed from the sample, eliminating the
concern that the results are driven by extreme observations.

99

3.5 Conclusion

How does US monetary policy spillover to international ﬁnancial markets? A common approach
in the literature is to use an event-study framework and ﬁrst-moment shocks (i.e. unexpected changes
in the expected path of the policy rate) to study this question. In this paper we argue that this typical
approach is not suﬃcient to capture the complete breadth of the international transmission of
US monetary policy actions. We show that changes in uncertainty around the Federal Reserve’s
expected policy path have important consequences for global bond and equity markets. Moreover,
omitting uncertainty from event-study regressions could lead to over-estimation of the ﬁrst moment
eﬀect, since changes in ﬁrst and second moments are positively correlated.

An increase in the market perceived uncertainty raises bond yields and lowers equity prices
in both advanced and emerging countries. However, the transmission works through diﬀerent
channels. The yield response in advanced countries is driven by changes in term premia and we
provide evidence for an international portfolio balance channel whereby bonds of countries that are
considered to be more substitutable vis-à-vis the US respond more to uncertainty. For emerging
countries, the yield response is driven by changes in the expected (or risk-neutral) component.
Using capital ﬂows data we show that uncertainty changes aﬀect capital outﬂows in a manner
consistent with a ﬂight to safety channel. Moreover, for emerging countries the responsiveness to
uncertainty is closely related to the country’s ﬁnancial openness.

Our results have implications for the design of monetary policy. We show that the uncertainty
spillover is substantially larger when the FOMC deliberately made changes to the forward guidance
language about future policy decisions. This suggests that the FOMC has an additional tool
for inﬂuencing international ﬁnancial conditions, namely by inﬂuencing the market’s perceived
uncertainty about the future path of the short rate. Moreover, in an environment where interest
rates are more likely to be constrained by the zero lower bound, changing uncertainty will likely
take on increasing importance in the FOMC’s toolkit.

Our work raises some natural questions that are worth exploring. We ﬁnd that the transmission
of US monetary policy uncertainty has gotten stronger since the ﬁnancial crisis, especially for

100

emerging countries. There is some evidence that an increase in ﬁnancial openness in emerging
countries has played a role in the higher responsiveness, but a more detailed analysis is warranted.
While we focus on the high-frequency response of ﬁnancial markets, evaluating the spillover eﬀects
of US monetary policy on lower frequency macroeconomic variables in advanced and emerging
countries appears to be a fruitful area for future research.

101

APPENDICES

102

APPENDIX A

SUPPLEMENTAL INFO FOR CHAPTER 1

Proof of Enforcement Constraint

The following proof of equation 1.10 follows the logic of Jermann & Quadrini (2012). After they
produce, sell output F(zt, kt, nt) and pay expenses, ﬁrms can then opt to default on their intraperiod
loan and renegotiate it. Thus, at the time of the default decision, ﬁrms are holding liabilities equal
to lt + bt+1
. Since all other expenses have been paid at this point, ﬁrms are holding liquidity exactly
1+rt
equal to lt + at+1 + (1 − ν)at, i.e. enough liquidity to pay the intraperiod loan, carry accumulated
liquidity to the next period and the amount of deferred labor expenses. Firms are also holding
non-liquid assets equal to kt+1, i.e. the physical capital. As in Jermann & Quadrini (2012), liquid
assets can be hidden from the lender; thus, the lender can only recoup physical capital.

In the event of default, the lender seizes the ﬁrm’s non-liquid assets and can liquidate them for
ξt ∗ kt+1. After the ﬁrm has decided to default, ξt is then revealed as either 0 or 1. Thus, the lender
will be able to either recoup the entire value of the physical capital or nothing.

If the ﬁrm decides to default, then the ﬁrm and lender enter a renegotiation process. For
simplicity, we assume that the ﬁrm has full bargaining power in the renegotiation, as changing the
bargaining power assumption for the renegotiation is equivalent to changing the value of ξt. Thus,
the formulation of the enforcement constraint (equation 1.10) is unaﬀected by this assumption. We
now consider the two extreme cases of ξt.

Case I: Lender recoups entire value of physical capital (ξ = 1)
In renegotiation, the ﬁrm must pay lender the amount kt+1 − bt+1
and promise to repay bt+1
1+rt
1+rt
next period. This is the amount that makes the lender indiﬀerent between liquidating the ﬁrm and
keeping the ﬁrm in operation. As discussed in the main text, in the event of default, the ﬁrm does
not have to pay back the intraperiod loan or its deferred labor costs. Thus, the ex-post value of

103

defaulting for the ﬁrm is:

Emt+1Vt+1 − kt+1 +

bt+1
1 + rt

. + lt + (1 − ν)at

(A.1)

Case II: Lender recoups nothing (ξ = 0)

In the event of ξt = 0, the lender will not want to liquidate the ﬁrm, as it cannot recoup anything
of value. The lender will simply choose to wait until next period when the ﬁrm will repay bt+1
.
1+rt
Thus, the ex-post value of defaulting for the ﬁrm is:

Emt+1Vt+1. + lt + (1 − ν)at

(A.2)

Since ξt is not revealed at the time lt is contracted, the expected value of default for the ﬁrm is:

Emt+1Vt+1. + lt + (1 − ν)at − ξt(kt+1 +

)

bt+1
1 + rt

(A.3)

In order for the lender to agree to intraperiod loan lt, the ﬁrm’s value of not defaulting

(Emt+1Vt+1) must be at least as high as the value of default:

Emt+1Vt+1 ≥ .Emt+1Vt+1. + lt + (1 − ν)at − ξt(kt+1 +

)

bt+1
1 + rt

Thus, we get our enforcement constraint:

ξt(kt+1 +

bt+1
1 + rt

) ≥ lt + (1 − ν)at = wtnt − νat

(A.4)

(A.5)

Alternative Household Budget Constraint

As mentioned above, we assume in the baseline model that the bank’s proﬁts, etlt, are immediately
consumed by the bank. Here, we distribute these proﬁts to the households as a lump-sum payment
and show that the results are essentially unchanged.
The household budget constraint now becomes

104

etlt + wtnt + bt + st(dt + pt) =

bt+1
1 + rt

+ st+1pt + ct + Tt,

where the bank’s proﬁts, etlt, have been added to the constraint. The rest of the budget constraint
remains the same.

Substituting in Equation (1.8) for etlt, nt = lt +at
wt

and lt = wtnt − at gives the new household

FOC for nt:

((1 + γ)ηwt + (1 − θ)(1 − η)zt k θ

t n−θ

t

)Uc(ct, nt) + Un(ct, nt) = 0

The household’s labor decision now inﬂuences household consumption via its eﬀect on the
bargaining problem between the ﬁrm and bank, i.e.
the total surplus generated from agreement
on the loan. The corresponding impulse response functions for our two steady states are shown in
Figure C9. The results closely match the baseline IRFs.

105

APPENDIX B

TABLES FOR CHAPTER 1

Table B1: Summary statistics

Panel A: Compustat

Small Firms

Large Firms

Assets (2012 $’s, millions)
Age (years)
Debt Issuance (% of assets)
Equity Issuance (% of assets)
Liquidity Accumulation (% of assets)
Debt-to-Assets Ratio
Bank Debt (% of total debt)
# of Firms

mean
71.5
10.8
0.7
19.1
5.9
31.4
18.6

std. dev.
112.1
9.4
21.9
59.8
37.0
50.3
32.0

mean
931.0
17.2
0.4
-0.2
1.1
33.9
27.6

std. dev.
961.0
13.1
15.8
11.3
11.7
31.5
33.6

13,158

3,517

Panel B: DealScan & Call Report

All Lender Pool

Lead Lender Pool

Lerner Index
Share of State Banking Assets
Number of Lenders
Recently Joined MBHC
Recently Acquired
Lead Year Share

Pre-1999 Post-1999

Pre-1999 Post-1999

0.476
0.129
2.425
0.181
0.040

0.535
0.230
3.374
0.122
0.080

0.476
0.129

0.185
0.041
0.892

0.536
0.252

0.119
0.080
0.792

Panel A displays summary statistics for the book value of assets, ﬁrm age and the key ﬁnancing variables.
The sample period is 1981-2017. Small ﬁrms are those with a book value of assets below the 60th percentile
in a given year. Large ﬁrms are those between the 60th percentile and 90th percentile. Panel B displays
means for characteristics of the lenders in the All Lender Pool and Lead Lender Pool. The sample period is
1985-2012.

106

Table B2: Cyclicality of Aggregate Financing Variables, by Size

Panel A: 1981-2017

Size categories Debt Iss
0.495***
Small Firms
Large Firms
0.622***
0.588***
All Firms

Equity Iss Liq. Accum.
0.276*
-0.430***
-0.086

0.180
-0.276*
-0.038

Panel B: 1981-1998

Size categories Debt Iss
0.556**
Small Firms
Large Firms
0.691***
0.648***
All Firms

Equity Iss Liq. Accum.
-0.487**
-0.557**
-0.484**

-0.367
-0.262
-0.323

Panel C: 1999-2017

Size categories Debt Iss
0.575**
Small Firms
Large Firms
0.653***
0.647***
All Firms

Equity Iss Liq. Accum.
0.509**
-0.414*
0.110

0.336
-0.307
0.061

This table displays the correlations between the cyclical component of HP-ﬁltered annual real corporate
GDP and the three ﬁnancing variables. The ﬁnancing variables are the cyclical component of the respective
HP-ﬁltered series, aggregated by the indicated ﬁrm size categories and normalized by the lagged book value
of assets. Small ﬁrms are those with book value of assets below the 60th percentile in a given year. Large
ﬁrms are those between the 60th percentile and 90th percentile. “All ﬁrms” are the pooled sample of small
and large ﬁrms. *** p < 0.01, ** p < 0.05, * p < 0.1

107

Table B3: Firm-Level Cyclicality of Financing Variables

Panel A: Baseline Speciﬁcation

1981-1998

1999-2017

Debt Iss. Equity Iss.

Liq. Accum.

Debt Iss. Equity Iss. Liq. Accum.

Small Firms

Large Firms

SF Observations
LF Observations

6.41***
(1.631)
8.80***
(1.436)

36,981
18,891

-10.74***
(3.569)
-2.92**
(1.195)

40,616
20,874

p-values

-8.61***
(2.964)
-2.51**
(1.158)

40,616
20,874

4.04***
(0.898)
6.08***
(1.323)

33,899
17,375

H0 : small = large

0.040

0.006

0.009

0.179

11.12**
(3.942)
-1.42**
(0.592)

39,363
19,698
p-values
0.003

3.56
(3.464)
-1.12
(1.081)

39,363
19,698

0.108

Panel B: With-In Firm Variance in Continuous Size Measure

1981-1998

1999-2017

Debt Iss. Equity Iss.

Liq. Accum.

Debt Iss. Equity Iss. Liq. Accum.

Cyclical GDP

Cyclical GDP x Size

6.59***
(1.430)
0.35
(0.919)

-6.52**
(2.312)
8.32***
(2.435)

-3.47
(2.257)
6.29***
(1.659)

4.61***
(0.629)
1.51*
(0.847)

6.95***
(2.377)
-8.62***
(2.288)

Observations

44,680

49,118

49,118

41,634

H0 : Interaction in Pre1999 = Post1999

0.927

47,965
p-values
0.000

2.95
(2.392)
-4.20**
(1.501)

47,965

0.000

This table displays the estimates of regressing the ﬁnancing variable of interest (as a percentage of ﬁrm
asset value) on the cyclical component of HP-ﬁltered annual real corporate GDP, normalized so that a unit
increase in GDP indicates moving from the lowest realization to the highest realization during the sample
period 1981-2017. Controls include the ﬁrm’s cash ﬂow and Tobin’s Q. Each coeﬃcient is the estimate
from a separate regression for each ﬁrm size x subperiod sample. Post-1999 estimates in bold indicate the
, where j ∈ {small, large}, is rejected at the 5% level. In Panel B, the GDP
hypothesis H0 : β
measure is interacted with a continuous measure of a ﬁrm’s book value of assets (Size). A ﬁrm-speciﬁc ﬁxed
eﬀect is included and all variables are demeaned by ﬁrm. Two-way clustered standard errors in parentheses,
*** p < 0.01, ** p < 0.05, * p < 0.1

post
j

pre
j

= β

108

Table B4: State-Level Timing of Riegle-Neal Adoption

Panel A: 1981-1998
Debt Iss. Equity Iss. Liq. Accum.

GDP

adopt1996 x GDP
adopt1997 x GDP

5.30***
(1.648)
2.50
(1.776)
1.01
(1.122)

-11.75***
(3.563)
-1.23
(2.497)
4.25
(2.871)

Observations

36,537

40,117

-9.63**
(3.324)
-0.24
(1.750)
4.34***
(1.274)

40,117

Panel B: 1999-2009
Debt Iss. Equity Iss. Liq. Accum.

GDP

adopt1996 x GDP
adopt1997 x GDP

5.08***
(0.554)
-1.54*
(0.722)
-0.76
(0.974)

11.89*
(5.502)
0.48
(3.359)
-7.91***
(2.458)

Observations

21,913

25,763

6.46
(5.377)
-2.16*
(1.101)
-6.75**
(2.498)

25,763

Panel C: 1999-2019
Debt Iss. Equity Iss. Liq. Accum.

GDP

adopt1996 x GDP
adopt1997 x GDP

4.86***
(0.829)
-1.11
(0.949)
-1.56
(1.156)

12.44***
(4.230)
-0.93
(2.274)
-6.02*
(2.996)

Observations

32,697

38,005

5.17
(3.907)
-1.12
(1.328)
-4.33*
(2.301)

38,005

This table displays the estimates of regressing the ﬁnancing variable of interest (as a percentage of ﬁrm
asset value) on the cyclical component of HP-ﬁltered annual real corporate GDP, normalized so that a unit
increase in GDP indicates moving from the lowest realization to the highest realization during the sample
period 1981-2017. This GDP measure is interacted with an indicator for the year of state-level Riegle-
Neal adoption. Controls include the ﬁrm’s cash ﬂow and Tobin’s Q. Two-way clustered standard errors in
parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

109

Table B5: Reduction in Small Firm’s Bargaining Power, Syndicate Structure

Panel A: Number of Lenders

1985-1998

1999-2012

Debt Iss. Equity Iss. Liq. Accum. Debt Iss. Equity Iss. Liq. Accum.

GDP

NumLenders x GDP

Observations
R2

0.99
(8.228)
2.57**
(0.858)
9,386
0.047

-8.41
(8.204)
-0.09
(1.245)
10,420
0.064

0.85
(7.076)
0.96
(0.650)
10,420
0.014

6.35***
(1.204)
2.62***
(0.530)
9,186
0.031

16.35*
(9.169)
-2.16*
(1.028)
10,761
0.026

3.25
(3.680)
-1.38**
(0.582)
10,761
0.005

Panel B: Lead Lender Share

1985-1998

1999-2012

Debt Iss. Equity Iss. Liq. Accum. Debt Iss. Equity Iss. Liq. Accum.

GDP

LeadShare x GDP

Observations
R2

4.40
(8.145)
-3.64*
(1.973)
9,334
0.039

-8.38
(8.011)
-0.61
(4.017)
10,356
0.062

4.64
(7.051)
-5.48
(3.279)
10,356
0.015

10.95***
(1.242)
-4.96***
(1.083)
9,015
0.025

10.59
(7.454)
7.27**
(2.505)
10,572
0.029

0.47
(3.028)
2.91*
(1.565)
10,572
0.005

This table displays the estimates of regressing the ﬁnancing variable of interest (as a percentage of ﬁrm asset
value) on GDP (i.e. the cyclical component of HP-ﬁltered real corporate GDP) and an interaction with the
number of lenders in the “All Lender Pool” (Panel A) or the percentage of a ﬁrm’s total syndicated loans
contributed by the lead lender(s) (Panel B) during the 1985-1998 and 1999-2012 period. Controls include
the ﬁrm’s cash ﬂow and Tobin’s Q. Two-way clustered standard errors in parentheses, *** p < 0.01, **
p < 0.05, * p < 0.1

110

Table B6: Reduction in Small Firm’s Bargaining Power, Lender Market Power

Panel A: Bank Merger, 1985-2012

All Lender Pool

Lead Lender Pool

Debt Iss. Equity Iss. Liq. Accum. Debt Iss. Equity Iss. Liq. Accum.
6.76***
(2.238)
0.26
(2.724)
18,572
0.007

6.66***
(2.145)
0.24
(3.451)
18,349
0.007

-0.01
(1.485)
6.44*
(3.234)
20,928
0.004

0.04
(1.500)
7.00**
(3.354)
21,181
0.004

7.00*
(3.443)
7.36**
(3.559)
20,928
0.020

7.28**
(3.419)
9.60**
(3.821)
21,181
0.020

Panel B: Size of Merger, 1999-2012

All Lender Pool

Lead Lender Pool

Debt Iss. Equity Iss. Liq. Accum. Debt Iss. Equity Iss. Liq. Accum.
8.99***
(1.242)
1.25
(2.506)
-3.23
(5.799)
9,091
0.017

13.66
(8.311)
5.06
(3.551)
23.38***
(4.974)
10,654
0.024

1.24
(3.279)
6.53
(3.982)
10.56***
(3.372)
10,654
0.004

8.93***
(1.274)
0.88
(2.034)
-11.07
(10.322)
8,892
0.016

1.12
(3.267)
5.36
(3.699)
11.41**
(3.984)
10,435
0.004

13.16
(8.326)
3.50
(3.667)
20.65**
(9.529)
10,435
0.024

GDP

Acquired x GDP

Observations
R2

GDP

Acquired x GDP

Size x Acquired x GDP

Observations
R2

GDP

Acquired x GDP

Panel C: By Size of Lender Pool, 1999-2012

Few Lenders

Many Lenders

Debt Iss. Equity Iss. Liq. Accum. Debt Iss. Equity Iss. Liq. Accum.
6.53***
(1.233)
1.75
(3.222)
5,641
0.010

12.89***
(2.893)
-5.34
(4.886)
3,374
0.033

18.29
(11.464)
8.13**
(3.676)
6,631
0.033

-2.00
(1.834)
0.44
(3.750)
3,941
0.004

3.39
(4.565)
6.59
(4.223)
6,631
0.006

4.15
(2.839)
-10.40
(7.914)
3,941
0.007

Observations
R2
This table displays the estimates of regressing the ﬁnancing variable of interest (as a percentage of ﬁrm
asset value) on cyclical GDP and an interaction with a ﬂag for a lender in a ﬁrm’s “All Lender Pool” or
“Lead Lender Pool” being acquired by another lender during the previous ﬁve years. Firms with few (many)
lenders are those with a below-average (above-average) number of lenders in their “All Lender Pool”. Size
is the percentage increase in the bank due to the merger. Controls include the ﬁrm’s cash ﬂow and Tobin’s
Q. Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

111

Table B7: Increase in Small Firms’ Lender Relationship Complexity, MBHC Status, 1999-2012

Panel A: Baseline Speciﬁcation

All Lender Pool

Lead Lender Pool

Debt Iss. Equity Iss. Liq. Accum. Debt Iss. Equity Iss. Liq. Accum.
9.63***
(1.144)
-4.32*
(2.196)
9,186
0.016

12.13
(7.925)
16.85**
(7.124)
10,761
0.025

9.55***
(1.233)
-5.26**
(1.788)
9,015
0.017

0.64
(3.089)
9.31
(5.471)
10,761
0.005

0.71
(3.093)
8.03
(5.123)
10,572
0.005

12.07
(7.999)
12.67*
(6.039)
10,572
0.025

Panel B: By Size of Lender Pool

Few Lenders

Many Lenders

Debt Iss. Equity Iss. Liq. Accum. Debt Iss. Equity Iss. Liq. Accum.
7.11***
(1.299)
-4.09*
(1.895)
5,641
0.010

13.26***
(2.834)
-7.57
(5.211)
3,374
0.033

16.93
(10.929)
17.58**
(6.745)
6,631
0.034

2.79
(4.266)
9.64
(5.923)
6,631
0.007

-2.10
(1.957)
1.44
(4.725)
3,941
0.005

3.96
(2.875)
-5.99
(6.152)
3,941
0.007

GDP

JoinMBHC x GDP

Observations
R2

GDP

JoinMBHC x GDP

Observations
R2

This table displays the estimates of regressing the ﬁnancing variable of interest (as a percentage of ﬁrm asset
value) on GDP (i.e. the cyclical component of HP-ﬁltered real corporate GDP) and an interaction with an
indicator for a lender in the “All Lender Pool” or the “Lead Lender Pool” that joined a multi-bank holding
company in the previous 5 years during the 1999-2012 period. Firms with few lenders are those with a
below-average number of lenders in their “All Lender Pool” and ﬁrms with many lenders are those with an
average or above number of lenders. Controls include the ﬁrm’s cash ﬂow and Tobin’s Q. Two-way clustered
standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

112

Table B8: Calibration

Discount Factor
Disutility of Work
Tax Advantage
Production Technology
Depreciation Rate
Payout cost parameter
Standard deviation: productivity shock σz = 0.006
σξ = 0.0087
Standard deviation: ﬁnancial shock
Value of liquidity as collateral
ν = 0.25

β = 0.9
α = 1.53
τ = 0.35
θ = 0.36
δ = 0.025
κ = 0.05

(cid:18)0.9736,−0.0287

(cid:19)

Matrix for the shocks process

0.1509, 0.9363

Steady State Values

Strong Relationship

Weak Relationship

Equity payout, d
Liquidity, a
Cost of intraperiod loan, e
Debt, b
Equity Issuance to output, ei/y
Debt issuance to output, di/y
Liquidity accumulation to output, la/y
Labor, n
Capital, k

0.076
0.001
0.017
1.745
-0.121
0.000
0.000
0.300
2.337

0.108
0.308
0.080
1.739
-0.173
0.000
0.000
0.300
2.316

This table displays the baseline parameter values and steady state values of the general equilibrium business
cycle model.

113

Table B9: Response of Financing Behavior to Positive TFP and Financial Shocks

Panel (a): Small Firms

Debt Iss.

Pre-1999 Post-1999

Equity Iss.

Pre-1999 Post-1999

Liq. Accum.

Pre-1999 Post-1999

TFP Shockt−1
Financial Shockt−1

Observations
R2

-0.45
(0.305)
1.05***
(0.289)

30,520
0.012

0.07
(0.197)
1.14***
(0.264)

33,899
0.005

0.42
(1.021)
-2.08***
(0.629)

33,780
0.081

2.43**
(1.057)
0.93
(1.919)

39,363
0.010

0.20
(0.928)
-1.50**
(0.649)

33,780
0.021

1.85**
(0.737)
-0.21
(1.549)

39,363
0.008

Panel (b): Large Firms

Debt Iss.

Pre-1999 Post-1999

Equity Iss.

Pre-1999 Post-1999

Liq. Accum.

Pre-1999 Post-1999

TFP Shockt−1
Financial Shockt−1

Observations
R2

-0.10
(0.294)
1.20***
(0.339)

14,406
0.026

0.02
(0.314)
1.83***
(0.526)

17,375
0.015

0.12
(0.314)
-0.60**
(0.275)

15,994
0.003

0.11
(0.103)
-0.50**
(0.183)

19,698
0.022

0.37
(0.233)
-0.59**
(0.270)

15,994
0.031

0.03
(0.203)
-0.35
(0.357)

19,698
0.008

This table displays the estimates of regressing the ﬁnancing variable of interest (as a percentage of ﬁrm asset
value) on the lagged annual value of the TFP shock and ﬁnancial shock. The shocks are standardized to
mean zero and unit variance. Controls include the ﬁrm’s cash ﬂow and Tobin’s Q. Panel (a) is the sample of
small ﬁrms and Panel (b) the sample of large ﬁrms. Small ﬁrms are those with book value of assets below
the 60th percentile in a given year. Large ﬁrms are those between the 60th percentile and 90th percentile.
Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

114

Table B10: Firm-Level Cyclicality of Real Variables

Panel A: Baseline Speciﬁcation

1981-1998

Investment

Employment

Investment Employment

1999-2017

Small Firms

Large Firms

SF Observations
LF Observations

-1.35
-0.886
0.94
-1.006

39,893
20,473

-1.86
-3.692
9.53**
-4.42

38,235
20,272

2.53***
-0.753
2.26**
-0.943

39,129
19,579

12.49***
-3.344
10.34***
-3.273

37,331
19,211

p-values

H0 : smallpre = smallpost
H0 : largepre = largepost

0.002
0.338

0.006
0.883

Panel B: By Liquidity Position, Small Firms 1981-2017

Low Liquidity Position
Employment

Investment

High Liquidity Position
Investment Employment

GDP

Dpost
t

-2.55***
0.004
5.05***
0.000

SF Observations
LF Observations

25,918
0.007

-7.32*
0.065

16.83***

0.001

24,973
0.014

-1.02
0.301
3.26***
0.007

24,700
0.011

2.83
0.452
8.72
0.119

23,988
0.029

This table displays the estimates of regressing change in investment (as a % of assets) and percentage change
in employment on the cyclical component of HP-ﬁltered annual real corporate GDP, normalized so that a unit
increase in GDP indicates moving from the lowest realization to the highest realization during the sample
period 1981-2017. Controls include the ﬁrm’s cash ﬂow and Tobin’s Q. Dpost
is an indicator for the years
1999-2017. Each coeﬃcient is the estimate from a separate regression for each ﬁrm size x subperiod sample.
In Panel B, Liquidity Position is determined by the median cash-to-assets ratio for the years 1996-1998.
Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

t

115

Table B11: Robustness Tests: Firm Fixed Eﬀect and Consistent Sample

Panel A: Firm Fixed Eﬀect

1981-1998

EI

DI

LA

DI

1999-2017

EI

LA

Small Firms

Large Firms

SF Observations
LF Observations

5.36***
(1.398)
7.40***
(1.264)

29,709
16,195

-11.37***
(3.413)
-4.61***
(0.957)

-6.56**
(2.981)
-2.59**
(1.052)

4.10***
(0.692)
6.43***
(1.079)

32,516
17,932

32,516
17,932

27,873
15,590

6.13
(3.649)
-1.57**
(0.550)

32,488
17,698

2.40
(3.270)
-1.35
(1.063)

32,488
17,698

Panel B: Consistent Sample

1981-1998

EI

DI

LA

DI

1999-2017

EI

LA

Small Firms

Large Firms

SF Observations
LF Observations

4.04*
(2.134)
4.90***
(1.631)

3,364
2,943

-3.65
(3.397)
-3.09***
(0.980)

-8.56**
(3.106)
-2.01
(1.310)

4.17**
(1.550)
5.30***
(0.871)

17.51**
(7.879)
-2.78***
(0.570)

7.04
(4.989)
-2.66**
(1.122)

3,677
3,227

3,677
3,227

4,051
3,603

4,627
4,042

4,627
4,042

This table displays the estimates of regressing the ﬁnancing variable of interest (as a percentage of ﬁrm
asset value) on the cyclical component of HP-ﬁltered annual real corporate GDP, normalized so that a unit
increase in GDP indicates moving from the lowest realization to the highest realization during the sample
period 1981-2017. In Panel A, a ﬁrm-speciﬁc ﬁxed eﬀect is included. In Panel B, the sample includes only
those ﬁrms that entered the Compustat sample prior to 1990 and also appeared in the sample in 2017. Controls
include the ﬁrm’s cash ﬂow and Tobin’s Q. Each coeﬃcient is the estimate from a separate regression for
post
,
each ﬁrm size x subperiod sample. 1999-2017 estimates in bold indicate the hypothesis H0 : β
j
where j ∈ {small, large}, is rejected at the 5% level. Two-way clustered standard errors in parentheses, ***
p < 0.01, ** p < 0.05, * p < 0.1

pre
j

= β

116

Table B12: Panel Regression: Alternative Financing Variables

Panel A: 1981-1998

Net Sale of Stock Alt. Liq. Accum.

Cash

Ret. Earnings

Small Firms

Large Firms

SF Observations
LF Observations

-10.82***
(3.564)
-2.96**
(1.147)

40,616
20,874

-4.63*
(2.575)
-1.40
(1.227)

35,736
17,153

p-values

-1.74
(2.278)
0.95
(1.304)

35,818
17,246

-9.37***
(1.737)
-0.89
(1.535)

40,616
20,874

H0 : small = large

0.006

0.083

0.087

0.001

Panel B: 1999-2017

Net Sale of Stock Alt. Liq. Accum.

Cash

Ret. Earnings

Small Firms

Large Firms

SF Observations
LF Observations

11.23**
(3.968)
-0.96
(0.600)

39,363
19,698

2.94
(2.331)
0.15
(0.567)

39,350
19,696

p-values

2.29
(2.268)
-0.15
(0.664)

39,237
19,510

-3.10
(3.291)
1.62
(2.063)

39,363
19,698

H0 : small = large

0.004

0.190

0.235

0.192

This table displays the estimates of regressing the ﬁnancing variable of interest (as a percentage of ﬁrm
asset value) on the cyclical component of HP-ﬁltered annual real corporate GDP, normalized so that a unit
increase in GDP indicates moving from the lowest realization to the highest realization during the sample
period 1981-2017. Controls include the ﬁrm’s cash ﬂow and Tobin’s Q. Each coeﬃcient is the estimate
from a separate regression for each ﬁrm size x subperiod sample. Panel B estimates in bold indicate the
, where j ∈ {small, large}, is rejected at the 5% level. Two-way clustered
hypothesis H0 : β
standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

post
j

pre
j

= β

117

Table B13: Aggregate Financing Variables: No Filtering

Size categories Debt Iss.
0.537***
Small Firms
Large Firms
0.431***
0.578***
All Firms

No Filtering
1981-2017
Equity Iss. Liq. Accum. Debt Iss.
0.625***
0.332**
0.098
0.602***
0.636***
0.279*

0.260
-0.161
0.191

Hamilton Filtering

1981-2017
Equity Iss. Liq. Accum.
0.404**
-0.233
0.140

0.328*
-0.160
0.141

Size categories Debt Iss.
0.575**
Small Firms
Large Firms
0.503**
0.652***
All Firms

1981-1998
Equity Iss. Liq. Accum. Debt Iss.
0.840***
0.210
-0.004
0.820***
0.320
0.838***

0.020
-0.199
0.069

1981-1998
Equity Iss. Liq. Accum.
-0.008
-0.476*
-0.226

0.022
-0.128
0.008

Size categories Debt Iss.
0.638***
Small Firms
Large Firms
0.554**
0.641***
All Firms

1999-2017
Equity Iss. Liq. Accum. Debt Iss.
0.639***
0.461**
-0.075
0.664***
0.690***
0.174

0.461**
-0.091
0.260

1999-2017
Equity Iss. Liq. Accum.
0.529**
-0.299
0.194

0.454*
-0.147
0.199

The left column of this table displays the correlations between the annual growth rate in real corporate GDP
and the three (non-HP-ﬁltered) ﬁnancing variables. The right column displays the correlations between the
cyclical component of Hamilton (2018)-ﬁltered annual real corporate GDP and the three ﬁnancing variables.
The ﬁnancing variables are aggregated by the indicated ﬁrm size categories and normalized by the lagged
book value of assets. Small ﬁrms are those with book value of assets below the 60th percentile in a given
year. Large ﬁrms are those between the 60th percentile and 90th percentile. “All ﬁrms” are the pooled
sample of small and large ﬁrms. *** p < 0.01, ** p < 0.05, * p < 0.1

118

Table B14: Eﬀects of Consolidation by Size of Lender Pool, 1999-2012, All Lender Pool

Many Lenders

Few Lenders

19.39
(11.488)

Debt Iss. Equity Iss. Liq. Accum. Debt Iss. Equity Iss. Liq. Accum.
7.00***
(1.203)
1.52
(3.294)
5,761
0.010

12.50***
(3.308)
-5.06
(8.910)
3,425
0.037

-2.31
(1.843)
0.93
(4.120)
3,998
0.006

3.69
(4.587)
6.05
(4.104)
6,763
0.006

3.60
(2.718)
-7.19
(8.455)
3,998
0.007

6.15
(3.664)
6,763
0.033

Few Lenders

Many Lenders

Debt Iss. Equity Iss. Liq. Accum. Debt Iss. Equity Iss. Liq. Accum.
7.52***
(1.276)
-3.59
(2.053)
5,761
0.011

12.94***
(2.893)
-6.34
(8.459)
3,425
0.032

17.96
(10.944)
17.25**
(7.267)
6,763
0.034

-2.39
(2.096)
1.62
(7.917)
3,998
0.006

3.03
(4.281)
10.12
(6.023)
6,763
0.007

2.62
(2.745)
5.23
(7.559)
3,998
0.006

GDP

Acquired x GDP

Observations
R2

GDP

JoinMBHC x GDP

Observations
R2

This table displays the estimates of regressing the ﬁnancing variable of interest (as a percentage of ﬁrm asset
value) on GDP (i.e. the cyclical component of HP-ﬁltered real corporate GDP) and an interaction with an
indicator for a lender in the “All Lender Pool” that was acquired by another lender during the previous ﬁve
years (Panel A) or that joined a multi-bank holding company in the previous 5 years (Panel B). Firms with
few lenders are those with a below-average number of lenders in their “All Lender Pool” and ﬁrms with
many lenders are those with an average or above number of lenders. Controls include the ﬁrm’s cash ﬂow
and Tobin’s Q. Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

119

Table B15: Panel Regression: Financing Response to Negative Shock

Panel A: 1981-1998

Debt Iss. Equity Iss. Liq. Accum.

Small Firms

Large Firms

SF Observations
LF Observations

-2.56***
(0.634)
-1.59*
(0.822)

36,981
18,891

H0 : small = large

0.007

-0.58
(1.505)
-0.12
(0.484)

40,616
20,874
p-values
0.678

0.89
(1.150)
0.55
(0.481)

40,616
20,874

0.659

Panel B: 1999-2017

Debt Iss. Equity Iss. Liq. Accum.

Small Firms

Large Firms

SF Observations
LF Observations

-1.74***
(0.487)
-1.63*
(0.937)

33,899
17,375

H0 : small = large

0.864

-7.02***
(2.236)
0.14
(0.340)

39,363
19,698
p-values
0.004

-4.92**
(1.800)
-0.39
(0.475)

39,363
19,698

0.008

This table displays the estimates of regressing the ﬁnancing variable of interest (as a percentage of ﬁrm asset
value) on an indicator for a “negative shock”, i.e. a year with negative growth in the cyclical component
of HP-ﬁltered real corporate GDP during the sample period 1981-2017. Years with a “negative shock” are
1982, 1986, 1989-1993, 2001-2003, 2007-2009, and 2016. Controls include the ﬁrm’s cash ﬂow and Tobin’s
Q. Each coeﬃcient estimate comes from running a separate regression on the ﬁrm size x subperiod sample.
, where j ∈ {small, large}, is rejected
Panel B estimates in bold indicate the hypothesis H0 : β
at the 5% level. Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

post
j

pre
j

= β

120

Table B16: Panel Regression: Real Response to Negative Shock

Panel A: 1981-1998

Investment Employment

Small Firms

Large Firms

SF Observations
LF Observations

Small Firms

Large Firms

SF Observations
LF Observations

-0.38
(0.422)
-0.86**
(0.359)

39,893
20,473

-4.10***
(1.197)
-5.67***
(1.325)

38,235
20,272

Panel B: 1999-2017

Investment Employment

-1.48***
(0.330)
-1.25***
(0.349)

39,129
19,579

-7.95***
(1.711)
-6.75***
(1.448)

37,331
19,211

p-values

H0 : smallpre = smallpost
H0 : largepre = largepost

0.045
0.437

0.070
0.580

This table displays the estimates of regressing change in investment (as a % of assets) and percentage change
in employment on an indicator for a “negative shock”, i.e. a year with negative growth in the cyclical
component of HP-ﬁltered real corporate GDP during the sample period 1981-2017. Years with a “negative
shock” are 1982, 1986, 1989-1993, 2001-2003, 2007-2009, and 2016. Controls include the ﬁrm’s cash
ﬂow and Tobin’s Q. Each coeﬃcient estimate comes from running a separate regression on the ﬁrm size x
subperiod sample. Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

121

APPENDIX C

FIGURES FOR CHAPTER 1

Figure C1: Concentration of US Banking Industry

This ﬁgure plots the national Herﬁndahl-Hirschman index for total loans and total assets in the
commercial banking sector during the period 1986-2017. Authors’ calculations using FR Y-9C
data.

122

Riegle-Neal0200400600800198019851990199520002005201020152020HHI - LoanHHI - AssetFigure C2: Time Series of Financial Variables, by Firm Size

(a) Debt Issuance and Cyclical GDP

(b) Equity Issuance and Cyclical GDP

(c) Liquidity Accumulation and Cyclical GDP

This ﬁgure plots the aggregate annual series of debt issuance (Panel a), equity issuance (Panel b)
and liquidity accumulation (Panel C) for small ﬁrms and large ﬁrms during the period 1980-2017.
RGDP (dotted line) is the cyclical component of HP-ﬁltered annual real corporate GDP. Small ﬁrms
are those with a book value of assets below the 60th percentile in a given year. Large ﬁrms are
those between the 60th percentile and 90th percentile. All series are standardized to have a mean
of zero and unit variance.

123

-4-2024198019851990199520002005201020152020YearSmallLargeRGDP-4-2024198019851990199520002005201020152020YearSmallLargeRGDP-4-2024198019851990199520002005201020152020SmallLargeRGDPFigure C3: Within-Period Model Timeline

124

Figure C4: IRFs of Financial Variables to Positive TFP & Financial Shocks

(a) Stronger Borrowers: High Bargaining Power / Low Bank Outside Option

(b) Weaker Borrowers: Low Bargaining Power / High Bank Outside Option

This ﬁgure plots the impulse responses of equity issuance (EI), liquidity accumulation (LA) and
debt issuance (DI) for a one standard deviation positive TFP shock (left column) and ﬁnancial shock
(right column). Panel (a) shows the impulse response when the ﬁrm bargaining power parameter
is set high and bank outside option is set low. Panel (b) shows the opposite. The y-axis is percent
deviation from the steady state value for the ratio of the ﬁnancing variable to output.

125

05101520-0.100.10.20.30.40.5TFP shockEI/YLA/YDI/Y05101520-0.5-0.4-0.3-0.2-0.100.10.20.30.40.5Financial shock05101520-0.100.10.20.30.40.5TFP shockEI/YLA/YDI/Y05101520-0.5-0.4-0.3-0.2-0.100.10.20.30.40.5Financial shockFigure C5: IRFs of Capital FOC Components to Positive TFP Shock

This ﬁgure plots the responses of strong borrowers (blue) and weak borrowers (red, dashed) to a
one standard deviation positive TFP shock. The y-axis is absolute deviation from the steady state
value.

126

05101520-0.0100.010.020.030.040.050.060.07Surplus Appropriation05101520-0.0100.010.020.030.040.050.060.07Financial Channel05101520-0.0100.010.020.030.040.050.060.07Collateral ChannelFigure C6: Wald Test for Structural Break in Cyclicality of Small Firm Financing

(a) Debt Issuance

(b) Equity Issuance

(c) Liquidity Accumulation

This ﬁgure plots the p-values of a Wald test to check for a structural break in the corresponding
year reported on the x-axis. Variables for debt issuance (Panel a), equity issuance (Panel b) and
liquidity accumulation (Panel c) are the aggregate series for small ﬁrms. Small ﬁrms are those with
book value of assets below the 60th percentile in a given year. The black horizontal line indicates
a p-value of 0.1.

127

Figure C7: Time Series of Financial & TFP Shocks

(a) Financial Shock & Cyclical GDP

(b) TFP Shock & Cyclical GDP

(c) Financial Shock & TFP Shock

This ﬁgure plots the model-implied annual series of the ﬁnancial shocks and TFP shocks during
the period 1984-2017. The dotted line in Panel (a) and Panel (b) is the cyclical component of
HP-ﬁltered annual real corporate GDP. Small ﬁrms are those with a book value of assets below
the 60th percentile in a given year. Large ﬁrms are those between the 60th percentile and 90th
percentile. All series are standardized to have a mean of zero and unit variance.

128

-4-2024198419881992199620002004200820122016YearFinancial ShockCyclical GDP-4-2024198419881992199620002004200820122016YearTFP ShockCyclical GDP-4-2024198419881992199620002004200820122016YearFinancial ShockTFP ShockFigure C8: IRFs of Capital FOC Components to Positive Financial Shock

This ﬁgure plots the responses of strong borrowers (blue) and weak borrowers (red, dashed) to a
one standard deviation positive ﬁnancial shock. The y-axis is absolute deviation from the steady
state value.

129

05101520-0.04-0.035-0.03-0.025-0.02-0.015-0.01-0.00500.0050.01Surplus Appropriation05101520-0.04-0.035-0.03-0.025-0.02-0.015-0.01-0.00500.0050.01Financial Channel05101520-0.04-0.035-0.03-0.025-0.02-0.015-0.01-0.00500.0050.01Collateral ChannelFigure C9: IRFs of Financial Variables: Alternative Household Budget Constraint

(a) Stronger Borrowers: High Bargaining Power / Low Bank Outside Option

(b) Weaker Borrowers: Low Bargaining Power / High Bank Outside Option

This ﬁgure plots the impulse responses of equity issuance (EI), liquidity accumulation (LA) and
debt issuance (DI) for a one standard deviation positive TFP shock (left column) and ﬁnancial shock
(right column). Panel (a) shows the impulse response when the ﬁrm bargaining power parameter
is set high and bank outside option is set low. Panel (b) shows the opposite. The y-axis is percent
deviation from the steady state value for the ratio of the ﬁnancing variable to output.

130

05101520-0.100.10.20.30.40.5TFP shockEI/YLA/YDI/Y05101520-0.5-0.4-0.3-0.2-0.100.10.20.30.40.5Financial shock05101520-0.100.10.20.30.40.5TFP shockEI/YLA/YDI/Y05101520-0.5-0.4-0.3-0.2-0.100.10.20.30.40.5Financial shockAPPENDIX D

TABLES FOR CHAPTER 2

Table D1: Summary Statistics

Pre-Crisis

Post-Crisis

Stock return
Leverage (Debt-to-Capital)
Leverage (Debt-to-Assets)
Leverage (Debt-to-Equity)
Implied volatility
MP shock
FFR shock
10 year shock
2 year shock

mean

0.240
0.381
0.251
0.985
34.799
-0.007
-0.020
0.001
-0.007

std. dev.
3.001
0.216
0.149
3.846
14.378
0.035
0.094
0.040
0.065

mean

0.162
0.414
0.276
1.208
27.846
-0.002
-0.003
-0.002
-0.006

std. dev.
1.935
0.231
0.162
3.538
9.453
0.023
0.012
0.035
0.035

The table shows summary statistics for stock returns, leverage measures, monetary policy shocks and
implied volatility. Stock returns and implied volatility are measured daily at the ﬁrm level. Leverage is
measured quarterly at the ﬁrm level. The monetary policy shocks are measured within a 30-minute window
around an FOMC announcement. Sample is non-ﬁnancial ﬁrms in S&P 500 on date of FOMC
announcement. Pre-crisis is Jul-1991 to Jun-2008 and post-crisis is Aug-2009 to Dec-2017.

131

Table D2: Response of ﬁrm-level stock returns to monetary shocks

Panel A:

(1a)

Pre-Crisis

(1b)

Post-Crisis

(2a)

Pre-Crisis

(2b)

Post-Crisis

(3a)

Pre-Crisis

(3b)

Post-Crisis

0.006
(0.039)

0.011
(0.027)

-0.026
(0.038)

0.006
(0.025)

0.006
(0.039)
-5.466*
(3.186)

0.009
(0.026)
2.223***
(0.573)

Leverage (Debt-to-Capital)

MP shock x Leverage

FFR shock x Leverage

10 yr shock x Leverage

2 yr shock x Leverage

-2.050*
(1.187)
0.265
(1.203)

-0.368
(1.382)
1.470***
(0.368)

Observations
R2
Firm controls
Firm FE
Time FE

48,143
0.181
yes
yes
yes

24,584
0.341
yes
yes
yes

48,143
0.181
yes
yes
yes

24,584
0.341
yes
yes
yes

Panel B:
Dpost
t
Dpost
t

x MP shock x Leverage (Debt-to-Capital)

x 2 yr shock x Leverage (Debt-to-Capital)

Observations
R2
Firm controls
Firm FE
Time FE

Full Sample

7.706**
(3.269)

72,733
0.206
yes
yes
yes

-1.205
(0.848)

48,143
0.177
yes
yes
yes

0.915**
(0.366)

24,584
0.341
yes
yes
yes

Full Sample

2.107**
(0.929)

72,733
0.203
yes
yes
yes

t + δli,t−1 + Γ(cid:48)Zi,t−1 + ei,t, where si,t is ﬁrm-level

Panel A shows results from estimating si,t = αi + αt + βli,t−1m
daily stock return, αi is a ﬁrm ﬁxed-eﬀect, αt is an FOMC day ﬁxed-eﬀect, li,t−1 is four-quarter moving average
leverage normalized to have mean 0 and variance 1, m
t
ﬁrm-level controls. The monetary policy shock is normalized to have a unit eﬀect on the 2 year yield and a positive
value represents an expansionary shock. Pre-crisis is Jul-1991 to Jun-2008 (153 obs.) and post-crisis is Aug-2009 to
Dec-2017 (68 obs.). Panel B shows results for
+ Γ(cid:48)Zi,t−1 + ei,t where Dpost
si,t = αi + αt + β1li,t−1m
t
indicator for the post-crisis period. Sample is non-ﬁnancial ﬁrms in S&P 500 on date of FOMC announcement.
Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

is the monetary policy shock and Zi,t−1 is a vector of

t + β2li,t−1m

t Dpost

+ δ1li,t−1 + δ2li,t−1Dpost

t

is an

t

132

Table D3: Regression of ﬁrm-level implied volatility leading up to FOMC announcement

Leverage

Post-Crisis x Leverage

Observations
R2
Firm FE
Time FE
Firm controls

(1)

(2)

-1.02**
(0.442)
2.30***
(0.403)

-0.75*
(0.414)
1.81***
(0.370)

47,131
0.759
yes
yes
no

42,635
0.786
yes
yes
yes

Null Hypothesis
leverage + post x leverage = 0

p-value

0.001

0.003

Results from estimating ivoli,t−1 = αi + αt + δli,t + βli,t−1Dpost
+ ΓZi,t−1 + ei,t, where ivoli,t−1 is
ﬁrm-level implied volatility on the day before the FOMC announcement, αi is a ﬁrm ﬁxed-eﬀect, αt is an
FOMC day ﬁxed-eﬀect, li,t−1 is four-quarter moving average leverage normalized to have mean 0 and
variance 1, Dpost
is an indicator for the post-crisis period and Zi,t−1 is the baseline vector of ﬁrm-level
controls including ﬁrm-level stock price at close of prior trading day. Pre-crisis is Jan-1996 to Jun-2008
(108 obs.) and post-crisis is Aug-2009 to Dec-2017 (68 obs.). Sample is non-ﬁnancial ﬁrms in S&P 500 on
date of FOMC announcement. Two-way clustered standard errors in parentheses, *** p < 0.01, **
p < 0.05, * p < 0.1

t

t

133

Table D4: Contemporaneous response of ﬁrm-level investment to monetary shocks

MP shockt x Leveraget−1
Dpost
t

x MP shockt x Leveraget−1

Observations
R2
Firm controls
Firm FE
Time FE

Investment
-3.71**
(1.596)
7.86***
(2.885)

19,755
0.146
yes
yes
yes

∆ln(yit) = αi + αt +n∈N β1nli,t−n−1m

Results from estimating

t−nDpost

t−n + Γ(cid:48)Zi,t−1 + eit, where yit is value of
t−n + β2nli,t−n−1m
ﬁrm i’s capital stock in quarter t, αi is a ﬁrm i ﬁxed eﬀect, αt is a quarter t ﬁxed eﬀect, lit is one-quarter
lagged leverage normalized to have mean 0 and variance 1, m
is the sum of all high-frequency monetary
t
is an indicator for the post-crisis period, N = [0, 12] and Zit−1
policy shocks that occur in quarter t, Dpost
is a vector of ﬁrm-level controls. The monetary policy shock is normalized to have a unit eﬀect on the 2
year yield and a positive value represents an expansionary shock. Sample is non-ﬁnancial S&P 500 ﬁrms
with at least 40 quarters of data in the pre-crisis or post-crisis sample for the dependent variable. Pre-crisis
is 1991:Q3 to 2008:Q2 and post-crisis is 2009:Q3 to 2017:Q4. Two-way clustered standard errors in
parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

t

Table D5: Response of Bond Yield Spread to Monetary Policy Shock

BAA - AAA spread

MP Shock

Post-Crisis x MP Shock

Observations
R2

1.034
(4.447)
-20.414*
(12.211)

221
0.021

t + βDpost
t
is a dummy for the post-crisis period and m
t

Results from estimating ∆ln(yt) = α0 + α1Dpost
yield spread, Dpost
monetary policy shock is normalized to have a unit eﬀect on the 2 year yield and a positive value represents
an expansionary shock. Pre-crisis is Jul-1991 to Jun-2008 (153 obs.) and post-crisis is Aug-2009 to
Dec-2017 (68 obs.). Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

m
t + eit, where yit is BAA-AAA bond
is the monetary policy shock. The

+ δm

t

t

134

Table D6: Response of 10 year nominal yield, real yield and term premium to monetary shocks

Panel A:

MP Shock

10 year nominal

10 year term premium
Pre-Crisis Post-Crisis Pre-Crisis Post-Crisis Pre-Crisis Post-Crisis
-1.45***
(0.323)

-1.77***
(0.319)

-1.89***
(0.415)

-0.40**
(0.157)

-0.19
(0.228)

10 year real

0.12
(0.168)

Observations
R2

153
0.012

68
0.311

83
0.104

68
0.352

153
0.010

68
0.309

Null Hypothesis
Dpost
t

x MP shock = 0

p-value
0.000

p-value
0.001

p-value
0.000

Panel B:

2 yr shock

10 year real

10 year nominal

10 year term premium
Pre-Crisis Post-Crisis Pre-Crisis Post-Crisis Pre-Crisis Post-Crisis
-0.63***
-0.42***
(0.077)
(0.203)

-0.71***
(0.244)

-0.21***
(0.051)

-0.33***
(0.066)

-1.01***
(0.216)

Observations
R2

153
0.205

68
0.121

83
0.226

68
0.248

153
0.103

68
0.142

Null Hypothesis
Dpost
t

x 2 yr shock = 0

p-value
0.268

p-value
0.003

p-value
0.044

t + eit, where yt is (daily) change in the 10 year nominal rate,
is the monetary policy

Results from estimating ∆yt = α0 + βm
10-year real rate, or the Kim & Wright 10 year term premium estimate and m
t
shock. The monetary policy shock is normalized so that a positive value represents an expansionary shock.
Pre-crisis is Jul-1991 to Jun-2008 (153 obs.) and post-crisis is Aug-2009 to Dec-2017 (68 obs.). The
10-year real rate is not available prior to 1999. The p-values for each panel are for a full sample estimation
of the interaction coeﬃcient Dpost
standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

is a dummy for the post-crisis period. Robust

, where Dpost

* m
t

t

t

135

Table D7: Response of ﬁrm-level stock returns (by high/low long-term leverage)

High LT Leverage

Low LT Leverage

Pre-Crisis Post-Crisis

Pre-Crisis Post-Crisis

(1a)

-2.738*
(1.475)

32,238
0.183
yes
yes
yes

(1b)

1.826**
(0.867)

18,817
0.350
yes
yes
yes

(2a)

-12.611*
(6.444)

15,901
0.231
yes
yes
yes

(2b)
0.245
(1.668)

5,766
0.342
yes
yes
yes

MP shock x Leverage

Observations
R2
Firm controls
Firm FE
Time FE

t + δli,t−1 + Γ(cid:48)Zi,t−1 + ei,t separately for ﬁrms classiﬁed
Results from estimating si,t = αi + αt + βli,t−1m
as “High (Low) LT Leverage" that are in the top two-thirds (bottom third) of the long-term (LT) leverage
distribution, where si,t is ﬁrm-level daily stock return, αi is a ﬁrm ﬁxed-eﬀect, αt is a time ﬁxed-eﬀect,
li,t−1 is four-quarter moving average leverage normalized to have mean 0 and variance 1, m
is the
t
monetary policy shock and Zi,t−1 is a vector of ﬁrm-level controls. The monetary policy shock is
normalized to have a unit eﬀect on the 2 year yield and a positive value represents an expansionary shock.
Pre-crisis is Jul-1991 to Jun-2008 (153 obs.) and post-crisis is Aug-2009 to Dec-2017 (68 obs.). Sample is
non-ﬁnancial ﬁrms in S&P 500 on date of FOMC announcement. Two-way clustered standard errors in
parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

Table D8: Response of ﬁrm-level stock returns to monetary policy uncertainty shocks

MPU Shock x Leverage

Observations
R2
Firm controls
Firm FE
Time FE

Pre-Crisis Post-Crisis

-0.117
(0.897)

39,684
0.187
yes
yes
yes

0.852*
(0.438)

24,584
0.341
yes
yes
yes

+ δli,t−1 + Γ(cid:48)Zi,t−1 + ei,t, where si,t is the ﬁrm-level
mpu
Results from estimating si,t = αi + αt + βli,t−1
t
daily stock return, αi is a ﬁrm ﬁxed-eﬀect, αt is an FOMC day ﬁxed-eﬀect, li,t−1 is 4-quarter moving
mpu
average leverage normalized to have mean 0 and variance 1, 
is monetary policy uncertainty shock
t
(positive value means a decrease in uncertainty) and Zi,t−1 is a vector of ﬁrm-level controls. Pre-crisis is
Feb-1994 to Jun-2008 (125 obs.) and post-crisis is Aug-2009 to Dec-2017 (68 obs.). Sample is
non-ﬁnancial ﬁrms in S&P 500 on date of FOMC announcement. Two-way clustered standard errors in
parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

136

Table D9: Response of Firm-Level Implied Volatility to MPU Shock

MPU Shock x Leverage

Post-Crisis x MPU Shock x Leverage

Observations
R2
Firm FE
Time FE
Firm controls

Full Sample

1.01**
(0.398)
-1.15***
(0.331)

9,965
0.367
yes
yes
yes

+ δlit−1 + Γ(cid:48)Zi,t−1 + ei,t,
mpu
Result from estimating ∆ivoli,t = αi + αt + β1li,t−1
t
where ivolit is the ﬁrm-level implied volatility, αi is a ﬁrm ﬁxed-eﬀect, αt is an FOMC day ﬁxed-eﬀect,
mpu
lit−1 is 4-quarter moving average leverage normalized to have mean 0 and variance 1, 
is monetary
t
policy uncertainty shock (positive value means a decrease in uncertainty)and Zit−1 is a vector of ﬁrm-level
controls. Pre-crisis is Jan-1996 to Jun-2008 (108 obs.) and post-crisis is Aug-2009 to Dec-2017 (68 obs.).
Sample is Liquid 100 non-ﬁnancial ﬁrms. Two-way clustered standard errors in parentheses, *** p < 0.01,
** p < 0.05, * p < 0.1

mpu
li,t−1
t

+ β2Dpost

t

Table D10: Summary Statistics of Firm Characteristics

Low Leverage

High Leverage

Leverage (Debt-to-Capital)
Firm age
Book value of assets ($, millions)
Market capitalization ($, millions)
Real sales growth (%, YoY)
Price-to-cost margin
Receivables-minus-payables to sales
Depreciation to assets
Current assets to total assets

mean

0.211
31.52
15,577
24,623
4.424
0.439
0.268
0.012
0.451

std. dev.
0.120
14.18
36,128
51,290
21.029
0.246
0.368
0.007
0.185

mean

0.562
37.62
21,613
18,898
2.010
0.353
0.245
0.011
0.298

std. dev.
0.146
13.03
47,465
37,891
21.858
0.221
0.677
0.006
0.172

The table shows summary statistics for the ﬁrm-level controls. The sample is divided into ﬁrms below
(“Low Leverage") and ﬁrms above (“High Leverage") the sample mean debt-to-capital ratio. All variables
are measured quarterly at the ﬁrm level. Sample is non-ﬁnancial ﬁrms in the S&P 500 between Jul-1991
and Dec-2017, excluding the ﬁnancial crisis dates of Jul-2008 to Jul-2009.

137

Table D11: Response of ﬁrm-level stock returns to monetary shocks w/ control interactions

MP shock x Leverage

MP shock x Current to total assets

MP shock x Real sales growth

MP shock x Firm size

MP shock x Price-to-cost margin

MP shock x Rec-minus-Pay to sales

MP shock x Depreciation-to-Assets

MP shock x Firm age

MP shock x Market capitalization

MP shock x 1st ﬁscal quarter

MP shock x 2nd ﬁscal quarter

MP shock x 3rd ﬁscal quarter

Observations
R-squared
Firm controls
Firm FE
Time FE

(1a)

Pre-Crisis

(1b)

Post-Crisis

-3.665**
(1.786)
13.676*
(8.058)
0.004
(0.044)
2.792
(1.769)
5.369
(6.225)
0.488
(1.040)
176.975
(177.487)
-0.236
(0.181)
-2.217
(1.636)
-3.209
(4.643)
-0.595
(2.199)
1.993
(3.344)

47,872
0.184
yes
yes
yes

1.001**
(0.437)
-4.722
(4.593)
0.021
(0.064)
0.776
(1.668)
2.314
(4.384)
-1.669**
(0.769)
-62.448
(122.824)

0.055
(0.050)
-1.204
(2.068)
2.206
(1.719)
-0.291
(2.362)
-1.774
(2.794)

24,516
0.343
yes
yes
yes

t + δlit−1 + Γ(cid:48)Zit−1 + eit, where sit is ﬁrm-level daily

Results from estimating sit = αi + αt + βlit−1m
stock return, αi is a ﬁrm ﬁxed-eﬀect, αt is an FOMC day ﬁxed-eﬀect, lit−1 is four-quarter moving average
leverage normalized to have mean 0 and variance 1, m
is the monetary policy shock and Zit−1 is a vector
t
of the baseline ﬁrm-level controls and ﬁrm’s sector (and their interactions with the MP shock). The
monetary policy shock is normalized to have a unit eﬀect on the 2 year yield and a positive value represents
an expansionary shock. Pre-crisis is Jul-1991 to Jun-2008 (153 obs.) and post-crisis is Aug-2009 to
Dec-2017 (68 obs.). Sample is non-ﬁnancial ﬁrms in S&P 500 on date of FOMC announcement. Two-way
clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

138

Table D12: Response of ﬁrm-level stock returns to monetary shocks: Without a time ﬁxed eﬀect

(1)

(2)

(3)

6.943*
(3.912)
-0.829
(7.333)

MP shock

Post-Crisis x MP shock

FFR shock

10 yr shock

Post-Crisis x FFR shock

Post-Crisis x 10 yr shock

2 yr shock

Post-Crisis x 2 yr shock

2.369
(1.990)
5.936***
(2.144)
6.896
(10.656)
-2.636
(4.593)

Observations
R2
Firm controls
Firm FE
Time FE

76,599
0.021
yes
yes
no

76,599
0.027
yes
yes
no

6.639***
(2.243)
7.483
(5.373)

76,599
0.038
yes
yes
no

t + δm

t Dpost

t

+ Γ(cid:48)Zit−1 + eit, where sit is ﬁrm-level daily stock return, αi is
Results from estimating sit = αi + βm
a ﬁrm ﬁxed-eﬀect, m
is an indicator for the post-crisis period and Zit−1 is a
t
vector of ﬁrm-level controls. The monetary policy shock is normalized to have a unit eﬀect on the 2 year yield and a
positive value represents an expansionary shock. Sample is Jul-1991 to Dec-2017 with post-crisis sample of
Aug-2009 to Dec-2017. Sample is non-ﬁnancial ﬁrms in S&P 500 on the date of FOMC announcement. Two-way
clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

is the monetary policy shock, Dpost

t

139

Table D13: Response of ﬁrm-level stock returns to monetary shocks (Pre. vs. Post 1SD leverage
outliers removed)

(1a)

Pre-Crisis

(1b)

Post-Crisis

(2a)

Pre-Crisis

(2b)

Post-Crisis

(3a)

Pre-Crisis

(3b)

Post-Crisis

0.015
(0.047)
-5.709*
(3.365)

0.001
(0.051)
4.428***
(1.010)

Leverage (Debt-to-Capital)

MP shock x Leverage

FFR shock x Leverage

10 yr shock x Leverage

2 yr shock x Leverage

0.015
(0.048)

0.001
(0.052)

-0.022
(0.051)

-0.013
(0.052)

-2.170*
(1.232)
-0.683
(1.241)

0.462
(1.078)
2.855***
(0.658)

-1.625
(0.999)

2.254***
(0.599)

22,731
0.205
yes
yes
yes

13,699
0.382
yes
yes
yes

22,731
0.205
yes
yes
yes

Observations
R-squared
Firm controls
Firm FE
Time FE

13,699
0.382
yes
yes
yes
t + δlit−1 + Γ(cid:48)Zit−1 + eit, where sit is ﬁrm-level daily stock return,
Results from estimating sit = αi + αt + βlit−1m
αi is a ﬁrm ﬁxed-eﬀect, αt is an FOMC day ﬁxed-eﬀect, lit−1 is four-quarter moving average leverage normalized to
have mean 0 and variance 1, m
is the monetary policy shock and Zit−1 is a vector of ﬁrm-level controls. The
t
monetary policy shock is normalized to have a unit eﬀect on the 2 year yield and a positive value represents an
expansionary shock. Pre-crisis is Jul-1991 to Jun-2008 (153 obs.) and post-crisis is Aug-2009 to Dec-2017 (68 obs.).
Sample is non-ﬁnancial ﬁrms in S&P 500 on date of FOMC announcement. We exclude 111 ﬁrms with a change in
leverage from pre-crisis to post-crisis greater than 1 standard deviation and 485 ﬁrms without an observation in either
the pre- or post-crisis sample. Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

22,731
0.202
yes
yes
yes

13,699
0.381
yes
yes
yes

140

Table D14: Robustness of baseline results to alternative measure of leverage: Debt-to-Assets

(1b)
(1a)
Firm Share Price

Pre-Crisis

Post-Crisis

(2)

Expected Volatility

Pre & Post

Leverage (Debt-to-Assets)

MP shock x Leverage

Dpost
t
Dpost
t

x Leverage

x MP shock x Leverage

0.004
(0.034)
-4.732*
(2.850)

-0.009
(0.026)
2.159***
(0.542)

Observations
R2
Firm controls
Firm FE
Time FE

48,169
0.180
yes
yes
yes

24,594
0.341
yes
yes
yes

-0.94**
(0.412)

1.87***
(0.365)

42,655
0.786
yes
yes
yes

(3)

Investment
Pre & Post

-5.59*
(3.014)
-12.94***
(4.315)
1.97
(4.710)
25.26***
(6.701)

19,441
0.147
yes
yes
yes

∆ln(yit) = αi + αt +

t + δlit−1 + Γ(cid:48)Zit−1 + eit, where sit is
Columns (1a) and (1b) are the results from estimating sit = αi + αt + βlit−1m
ﬁrm-level daily stock return, αi is a ﬁrm ﬁxed-eﬀect, αt is an FOMC day ﬁxed-eﬀect, lit−1 is four-quarter moving
average debt-to-assets normalized to have mean 0 and variance 1 , m
is the monetary policy shock and Zit−1 is a
t
vector of ﬁrm-level controls. Column (2) is the result from estimating
ivoli,t−1 = αi + αt + δli,t + βli,t−1Dpost
n∈N β1nli,t−n−1m

t−n + Γ(cid:48)Zi,t−1 + eit, where yit is the value of
ﬁrm i’s capital stock in quarter t. The monetary policy shock is normalized to have a unit eﬀect on the 2 year yield
and a positive value represents an expansionary shock. The pre-crisis sample is Jul-1991 to Jun-2008 (153 obs.) and
post-crisis is Aug-2009 to Dec-2017 (68 obs.). Sample is non-ﬁnancial ﬁrms in S&P 500 on date of FOMC
announcement. Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

+ ΓZi,t−1 + ei,t. Column (3) is the result from estimating
t−n + β2nli,t−n−1m

t−nDpost

t

141

Table D15: Robustness of baseline results to alternative measure of leverage: 1-quarter lagged
debt-to-capital

(1a)
(1b)
Firm Share Price

Pre-Crisis

Post-Crisis

(2)

Expected Volatility

Pre & Post

Leverage (Debt-to-Capital)

MP shock x Leverage

Dpost
t
Dpost
t

x Leverage

x MP shock x Leverage

0.014
(0.040)
-4.990
(3.058)

0.012
(0.029)
2.127***
(0.599)

Observations
R2
Firm controls
Firm FE
Time FE

48,895
0.181
yes
yes
yes

24,928
0.341
yes
yes
yes

-0.58
(0.385)

1.75***
(0.362)

43,255
0.786
yes
yes
yes

(3)

Investment
Pre & Post

0.88*
(0.483)
-3.78**
(1.601)
-1.02
(0.656)
7.12**
(2.804)

18,488
0.160
yes
yes
yes

∆ln(yit) = αi + αjt +

t + δlit−1 + Γ(cid:48)Zit−1 + eit, where sit is
Columns (1a) and (1b) are the results from estimating sit = αi + αt + βlit−1m
the ﬁrm-level daily stock return, αi is a ﬁrm ﬁxed-eﬀect, αt is an FOMC day ﬁxed-eﬀect, lit−1 is one-quarter lagged
leverage normalized to have mean 0 and variance 1, m
t
ﬁrm-level controls. Column (2) is the result from estimating
ivoli,t−1 = αi + αt + δli,t + βli,t−1Dpost

t−n + Γ(cid:48)Zi,t−1 + eit, where yit is the value of
ﬁrm i’s capital stock in quarter t. The monetary policy shock is normalized to have a unit eﬀect on the 2 year yield
and a positive value represents an expansionary shock. Pre-crisis is Jul-1991 to Jun-2008 (153 obs.) and post-crisis is
Aug-2009 to Dec-2017 (68 obs.). Sample is non-ﬁnancial ﬁrms in S&P 500 on date of FOMC announcement.
Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

+ ΓZi,t−1 + ei,t. Column (3) is the result from estimating
t−n + β2nli,t−n−1m

t−nDpost

is the monetary policy shock and Zit−1 is a vector of

t

n∈N β1nli,t−n−1m

142

Table D16: Robustness of baseline results to time-sector FE

(1a)
(1b)
Firm Share Price

Pre-Crisis

Post-Crisis

Leverage (Debt-to-Capital)

MP shock x Leverage

Dpost
t
Dpost
t

x Leverage

x MP shock x Leverage

0.001
(0.037)
-4.628*
(2.762)

0.007
(0.027)
1.461***
(0.549)

Observations
R2
Firm controls
Firm FE
Time-Sector FE

47,737
0.225
yes
yes
yes

24,450
0.401
yes
yes
yes

(2)

Expected Volatility

Pre & Post

-0.58
(0.408)

1.90***
(0.368)

(3)

Investment
Pre & Post

-2.08*
(1.114)
-3.94**
(1.718)
2.97
(2.616)
7.02**
(3.212)

19,323
0.181
yes
yes
yes

42,468
0.810
yes
yes
yes
t + δlit−1 + Γ(cid:48)Zit−1 + eit, where sit
is the monetary policy shock and

∆ln(yit) = αi + αjt +

Columns (1a) and (1b) are the results from estimating sit = αi + αjt + βlit−1m
is ﬁrm-level daily stock return, αi is a ﬁrm ﬁxed-eﬀect, αjt is a sector j by FOMC day ﬁxed-eﬀect, lit−1 is
four-quarter moving average leverage normalized to have mean 0 and variance 1, m
t
Zit−1 is a vector of ﬁrm-level controls. Column (2) is the result from estimating
ivoli,t−1 = αi + αjt + δli,t + βli,t−1Dpost
n∈N β1nli,t−n−1m

+ ΓZi,t−1 + ei,t. Column (3) is the result from estimating
t−n + β2nli,t−n−1m

t−n + Γ(cid:48)Zi,t−1 + eit, where yit is the value of
ﬁrm i’s capital stock in quarter t. The monetary policy shock is normalized to have a unit eﬀect on the 2 year yield
and a positive value represents an expansionary shock. Pre-crisis is Jul-1991 to Jun-2008 (153 obs.) and post-crisis is
Aug-2009 to Dec-2017 (68 obs.). Sample is non-ﬁnancial ﬁrms in S&P 500 on date of FOMC announcement.
Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

t−nDpost

t

143

Table D17: Robustness of baseline results to full CRSP/Compustat sample: No ﬁrm entry/exit

(1a)
(1b)
Firm Share Price

Pre-Crisis

Post-Crisis

-0.004
(0.038)
-2.336**
(1.071)

0.046
(0.054)
1.905**
(0.825)

Leverage (Debt-to-Capital)

MP shock x Leverage

Dpost
t
Dpost
t

x Leverage

x MP shock x Leverage

Observations
R2
Firm controls
Firm FE
Time FE

75,545
0.081
yes
yes
yes

38,324
0.232
yes
yes
yes

(2)

Investment
Pre & Post

-1.06***
(0.258)
-1.66
(1.034)
-0.25
(0.390)
3.23*
(1.747)

78,665
0.087
yes
yes
yes

∆ln(yit) = αi + αjt +

t + δlit−1 + Γ(cid:48)Zit−1 + eit, where sit is
Columns (1a) and (1b) are the results from estimating sit = αi + αt + βlit−1m
ﬁrm-level daily stock return, αi is a ﬁrm ﬁxed-eﬀect, αt is an FOMC day ﬁxed-eﬀect, lit−1 is four-quarter moving
average leverage normalized to have mean 0 and variance 1, m
is the monetary policy shock and Zit−1 is a vector of
t
ﬁrm-level controls. Column (2) is the result from estimating
t−n + Γ(cid:48)Zi,t−1 + eit, where yit is the value of
t−nDpost
ﬁrm i’s capital stock in quarter t. The monetary policy shock is normalized to have a unit eﬀect on the 2 year yield
and a positive value represents an expansionary shock. Pre-crisis is Jul-1991 to Jun-2008 (153 obs.) and post-crisis is
Aug-2009 to Dec-2017 (68 obs.). Sample is non-ﬁnancial ﬁrms in the CRSP/Compustat sample for the entire sample
period. Two-way clustered standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1

n∈N β1nli,t−n−1m

t−n + β2nli,t−n−1m

144

APPENDIX E

FIGURES FOR CHAPTER 2

Figure E1: Firm Leverage Plots

(a) Distribution of Firm Leverage

(b) Firm Leverage: Post vs. Pre

(c) Conditional Correlation of Credit Rating and
Leverage: Pre-Crisis

(d) Conditional Correlation of Credit Rating and
Leverage: Post-Crisis

Panel (a) plots the histogram of the quarterly ﬁrm leverage (measured as debt-to-capital), averaged
across the pre-crisis (grey, shaded) and post-crisis (red, transparent) samples. Panel (b) plots the
scatter plot of quarterly ﬁrm leverage (measured as debt-to-capital) averaged across the post-crisis
versus the average in the pre-crisis sample. Firms further than one standard deviation from the
45-degree line are shown in red. Panels (c) and (d) plot the residuals from regressing ﬁrm’s S&P
long-term credit rating on our set of control variables against the residuals from regressing ﬁrm’s
4-qtr rolling leverage on a set of control variables.

145

Figure E2: Eﬀects of an expansionary monetary policy shock

146

MC0MC1MB0MB1k0k0⇤1k0⇤0Lower-LeverageFirm:Pre-CrisisMC0MC1MB0MB1k0k0⇤1k0⇤0Higher-LeverageFirm:Pre-CrisisMC0MC1MB0MB1k0k0⇤1k0⇤0Lower-LeverageFirm:Post-CrisisMC0MC1MB0MB1k0k0⇤1k0⇤0Higher-LeverageFirm:Post-CrisisAPPENDIX F

SUPPLEMENTAL INFO FOR CHAPTER 3

F.0.1

Interpretation of monetary policy uncertainty

In this section we discuss the interpretation of the monetary policy uncertainty from the perspective
of a simple policy rule. Consider a general monetary policy rule of the ofrm

it = g (Ωt) + εt

(F.1)

where it is the short-term nominal interest rate, g (.) is a linear function that depends on the central
bank’s information set (Ωt) and εt is the monetary policy shock. In this framework uncertainty
about the interest rates comes from i)variance of εt and ii)any uncertainty about reaction function
g (Ωt). Speciﬁcally the h period ahead variance of the interest rate can be written as

Vart[it+h] = Vart[g (Ωt+h)] + Vart[εt+h] + covariance terms

(F.2)

What would cause changes in Vart[εt+h]? As discussed in the survey on monetary policy
shocks (Christiano et al. (1999)), there are various interpretations of what constitutes the shock ε
itself. For example, the shock can arise due to changing political pressure on the Fed, changing
composition of the voting members of the FOMC, technical factors like measurement error in the
preliminary data available to the FOMC when it makes its decision or even strategic aspects as
FOMC wanting to avoid disappointing market expectations. Any change in the expected importance
of these factors would lead to a change in Vart[εt+h].

What could cause changes in Vart[g (Ωt+h)]? Any uncertainty about future changes to the
reaction function will drive this. For example, consider a simple version of the Taylor rule
g (Ωt+h) = r∗ + π∗ + αyt + βπt. Uncertainty about the parameters (α, β), inﬂation target (π∗) or
equilibrium real interest rate (r∗) will drive Vart[g (Ωt+h)]. 1
1Note that short-rate uncertainty is also driven by uncertainty about economic fundamentals in addition to the two
In other words, in the above example Vart[πt+h] and
sources of monetary policy uncertainty discussed above.

147

Thus changes in monetary policy uncertainty around FOMC announcements can be interpreted
as changes in uncertainty about the monetary reaction function or changes in the variance of the
shock to the policy rule.

F.0.2 Construction of monetary policy uncertainty measure (mpu)

Our measure is based on the approach of Bauer et al. (2019) which uses options on Eurodollar futures
and does not require distributional assumptions. Let Ft,T be the time-t value of a Eurodollar futures
contract expiring at T with value at expiration is FT,T = 100 − LT, where LT is LIBOR in percent.
For option contracts, denote the payoﬀ max(FT,T − K, 0) for call options and max(K − FT,T, 0) for
put options, where K is the strike price. For a given trading date t and an expiration date T we use the
prices of call options, ct,T(K), and put options, pt,T(K) to calculate the market-based conditional
variance of future LIBOR, Vart(LT). Starting from the relationship Vart(LT) = Vart(FT,T) =
Et F2

T,T − (Et FT,T)2 = Et F2

t,T, we can show that2

T,T − F2
Vart(LT) =

Þ ∞
(cid:32)Þ Ft,T
(cid:20) ct,T(K)
Þ ∞

2
Pt,T
2
Pt,T

0

0

0

Pt,T

=

= 2

ct,T(K)dK − F2

t,T

Þ ∞

Ft,T

pt,T(K) +

ct,T(K)dK

− max(0, Ft,T − K)

dK

(cid:33)

(cid:21)

We then approximate this integral with out-of-the money option prices. While Eurodollar contracts
have a ﬁxed maturity, we interpolate contracts to get a ﬁxed-horizon measure. Our baseline

measure mpu is the two-day change around FOMC announcements in(cid:112)Vart(Lt+h) with h equal

to 12 months.

Vart[yt+h] will also aﬀect Vart[it+h]. Fortunately, a decomposition of high-frequency uncertainty changes coming
from macro variables and that coming from monetary policy (shock and reaction function) is not necessary for
the purpose of our paper. Our solution to identifying changes to monetary policy uncertainty is to follow a large
event-study literature and squarely focus on the changes in asset prices and our uncertainty measure over short time
windows containing FOMC announcements. Over these short time windows, asset prices are driven by the news in
these announcements. Since these changes in uncertainty are due to monetary policy announcements, we speak of
monetary policy uncertainty.
2For details on the derivation see Bauer et al. (2019).

148

APPENDIX G

TABLES FOR CHAPTER 3

Table G1: Summary Statistics

Mean Median Std Dev Min Max Observations

Panel (a): US monetary policy shocks
mps
mpu
Panel (b): International asset prices
2 year yield

0.00
-0.49

0.08
-0.39

0.91
1.00

0.09
0.16

Advanced
Emerging
10 year yield
Advanced
Emerging

Stock return

Advanced
Emerging
Exchange rate
Advanced
Emerging

-0.01
-0.01

-0.01
0.00

-0.01
-0.02

-0.01
-0.01

0.19
0.29

0.03
0.02

0.20
0.17

0.01
0.00

0.08
0.18

1.98
2.40

0.93
1.20

-4.37
-4.96

3.08
2.07

204
204

-1.28
-3.14

-0.98
-2.19

-12.21
-18.41

-10.05
-13.74

2.04
1.04

0.41
0.81

13.86
18.41

7.10
30.76

4,154
1,270

4,154
1,270

5,129
3,102

5,709
3,130

Panel (a) shows summary statistics for the monetary policy surprise (mps) and monetary policy uncertainty
(mpu) shock measures calculated in a two day window around FOMC announcements. Panel (b) shows
summary statistics for changes in 2 and 10 year government bond yields, stock returns, and exchange rate
returns (foreign currency relative to US dollar) for the countries in our sample. All changes and returns
are calculated in a two day window around FOMC announcements. We have government yield data for
22 advanced countries and 8 emerging countries. For exchange rates and stock prices we have data for 28
advanced countries and 16 emerging countries, see Table G13 for details.

149

Table G2: Response of US asset prices to monetary shocks

Jan-1995 to Jun-2019

2 year yield

10 year yield

S&P 500

mps

mpu

Constant

0.882***
[0.097]

-0.112***
[0.042]

0.876***
[0.104]
0.013
[0.069]
-0.106*
[0.056]

Observations
R-squared

204
0.646

204
0.646

0.548***
[0.106]

-0.087
[0.061]

204
0.249

0.382***
[0.095]
0.365***
[0.081]
0.093
[0.070]

204
0.359

-0.179*
[0.093]

0.161**
[0.069]

204
0.027

-0.137
[0.096]
-0.091
[0.078]
0.116
[0.085]

204
0.033

Jan-1995 to Nov-2007

mps

mpu

Constant

Observations
R-squared

2 year yield

10 year yield

S&P 500

0.911***
[0.087]

-0.059
[0.051]

108
0.724

0.874***
[0.102]
0.096
[0.072]
-0.012
[0.063]

108
0.732

0.573***
[0.119]

-0.056
[0.082]

108
0.286

0.450***
[0.125]
0.322***
[0.113]
0.100
[0.100]

-0.181
[0.139]

0.314***
[0.095]

-0.216
[0.141]
0.091
[0.099]
0.358***
[0.112]

108
0.377

108
0.029

108
0.036

Dec-2007 to Jun-2019

2 year yield

10 year yield

S&P 500

mps

mpu

Constant

0.840***
[0.165]

-0.203***
[0.069]

0.875***
[0.170]
-0.065
[0.110]
-0.236***
[0.090]

Observations
R-squared

96
0.552

96
0.555

0.547***
[0.194]

-0.115
[0.090]

96
0.234

0.331**
[0.158]
0.395***
[0.116]
0.085
[0.100]

96
0.353

-0.212*
[0.123]

0.018
[0.101]

96
0.035

-0.070
[0.123]
-0.259**
[0.110]
-0.113
[0.116]

96
0.087

The table shows the response of 2 and 10 year Treasury bond yields and the S&P 500 to a monetary policy
surprise and monetary policy uncertainty shock. All variables have been normalized to have unit standard
deviation. The full sample consists of 204 FOMC announcements from January 1995 to June 2019, the
pre-crisis sample consists of 108 FOMC announcements from January 1995 to November 2007, and the
post-crisis sample consists of 96 FOMC announcements from December 2007 to June 2019. All changes are
calculated in a two day window around FOMC announcements. Heteroskedasticity-robust standard errors
are reported in parentheses.

150

Table G3: Response of international bond yields to monetary shocks

Advanced countries

2 year yield
(1)

(2)

10 year yield
(1)
(2)

0.410***
[0.065]

-0.103***
[0.036]

0.360***
[0.061]
0.107**
[0.048]
-0.051
[0.042]

0.345***
[0.080]

-0.115**
[0.048]

0.221***
[0.069]
0.264***
[0.059]
0.014
[0.054]

mps

mpu

Constant

Observations
R-squared

4,154
0.137

4,154
0.146

4,154
0.097

4,154
0.154

Emerging countries

2 year yield
(1)

(2)

0.238***
[0.050]

-0.077**
[0.031]

0.159***
[0.043]
0.158***
[0.038]
-0.000
[0.033]

10 year yield
(1)
(2)

0.214**
[0.070]

-0.108***
[0.028]

0.129*
[0.056]
0.171***
[0.045]
-0.025
[0.030]

mps

mpu

Constant

Observations
R-squared

1,270
0.044

1,270
0.065

1,270
0.036

1,270
0.060

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps)
and monetary policy uncertainty (mpu) shock. All variables have been normalized to have unit standard
deviation. Column (1) has only mps as a regressor, while column 2 adds mpu to this speciﬁcation. The
sample consists of 204 FOMC announcements from January 1995 to June 2019. All changes are calculated
in a two day window around FOMC announcements. Standard errors reported in parentheses are calculated
with two-way clustering (along the country and time dimension).

151

Table G4: Response of international bond yields to monetary and macro data news shocks

Advanced countries

Emerging Countries

2 year yield
0.382***
[0.066]
0.111**
[0.051]
4,154
0.145

10 year yield

0.239***
[0.074]
0.283***
[0.065]
4,154
0.155

2 year yield
0.163***
[0.045]
0.174***
[0.048]
1,270
0.069

10 year yield

0.127**
[0.053]
0.179***
[0.049]
1,270
0.055

-0.004
[0.019]
5,562
0.000

0.023
[0.017]
1,900
0.008

0.017
[0.017]
5,562
0.001

0.010
[0.017]
1,900
0.002

0.099***
[0.029]
5,583
0.045

0.064***
[0.020]
5,583
0.019

0.037*
[0.019]
5,583
0.007

0.016
[0.017]
5,495
0.001

0.007
[0.021]
5,583
0.000

0.017
[0.020]
5,495
0.001

-0.010
[0.010]
1,681
0.000

0.018
[0.017]

573
0.004

0.019
[0.013]
1,685
0.001

0.009
[0.013]
1,685
0.000

0.007
[0.013]
1,654
0.000

-0.007
[0.012]
1,681
0.000

-0.003
[0.014]

573
0.000

0.005
[0.015]
1,685
0.000

0.012
[0.016]
1,685
0.001

-0.004
[0.020]
1,654
0.000

mps

mpu

Observations
R-squared

Unemployment

Observations
R-squared

GDP

Observations
R-squared

Retail Sales

Observations
R-squared

CPI

Observations
R-squared

PPI

Observations
R-squared

The table shows the response of 2 and 10 year government bond yields in advanced and emerging countries
to a monetary policy surprise (mps), monetary policy uncertainty (mpu) and news shocks. All variables
have been normalized to have unit standard deviation. For the employment report, we use non-farm payrolls,
for CPI and PPI we use headline inﬂation, retail sales are the total sales including automobiles, GDP is
the advance GDP release. The sample runs from January 1995 to June 2019. Standard errors reported in
parentheses are calculated with two-way clustering (along the country and time dimension).

152

Table G5: Response of expected component and term premium component of international bond
yields to monetary shocks

Panel (a)

mpu

Advanced countries
10y ec
-0.004
[0.052]

2y tp
0.129**
[0.048]

2y ec
-0.012
[0.049]

10y tp
0.241***
[0.060]

2y ec
0.094*
[0.044]

Emerging countries
2y tp
10y ec
0.070
0.128***
[0.034]
[0.039]

10y tp
0.056
[0.040]

Panel (b)

mpu

US 10y tp

Advanced countries
10y ec
0.040
[0.053]
-0.100*
[0.051]

2y tp
0.037
[0.035]
0.210***
[0.044]

10y tp
0.067
[0.053]
0.402***
[0.046]

2y ec
0.041
[0.045]
-0.122**
[0.048]

Emerging countries
10y ec
2y tp
0.044
0.124***
[0.035]
[0.030]
0.008
0.058*
[0.029]
[0.029]

10y tp
0.036
[0.046]
0.043
[0.041]

2y ec
0.109*
[0.047]
-0.034
[0.031]

Panel (a) shows the response of the expected component (ec) and term premium (tp) of 2 and 10 year
government bond yields to a monetary policy uncertainty (mpu) shock. The monetary policy surprise (mps)
and a constant are included in the regressions, the coeﬃcients are left out for space considerations. Yields are
decomposed into the expected component and term premium using the methodology of Joslin et al. (2011).
All variables have been normalized to have unit standard deviation. Panel (b) adds the US 10 year yield
term premium to the speciﬁcation. The sample consists of 204 FOMC announcements from January 1995 to
June 2019. All changes are calculated in a two day window around FOMC announcements. Standard errors
reported in parentheses are calculated with two-way clustering (along the country and time dimension).

153

Table G6: Response of exchange rates to monetary shocks

Advanced

(1)

(2)

0.240***
[0.064]

0.034
[0.049]

0.281***
[0.067]
-0.091
[0.062]
-0.011
[0.058]

mps

mpu

Constant

Observations
R-squared

5,709
0.048

5,709
0.054

Emerging

(1)

0.130***
[0.044]

0.008
[0.035]

3,130
0.014

(2)

0.085*
[0.041]
0.098**
[0.044]
0.056
[0.034]

3,130
0.022

The table shows the response of international exchange rate returns to a monetary policy surprise (mps)
and monetary policy uncertainty (mpu) shock. The sample consists of 204 FOMC announcements from
January 1995 to June 2019. Exchange rate returns have been normalized to have unit standard deviation.
Exchange rates are in units of foreign currency per US dollar such that an increase represents a depreciation
of the foreign currency relative to the dollar. All changes are calculated in a two day window around FOMC
announcements. Standard errors reported in parentheses are calculated with two-way clustering (along the
country and time dimension).

Table G7: Response of term premium component of international bond yields to monetary shocks
(bond substitutability interaction)

10 year term premium

All Countries Advanced Emerging

-0.33***
(0.080)
0.72***
(0.148)

5,410
0.045

-0.16***
(0.044)
0.53***
(0.077)

4,140
0.062

-0.12
(0.302)
0.29
(0.526)

1,270
0.008

mpu

mpu x bond sub.

Observations
R-squared

The table shows the response of 10 year government bond yield term premia to a monetary policy uncertainty
(mpu) shock and the interaction with a measure of bond substitutability with the United States. The monetary
policy surprise (mps), its interaction with bond substitutability and a constant are included in the regressions,
the coeﬃcients are left out for space considerations. The term premium is calculated using the methodology
of Joslin et al. (2011). Bond substitutability is calculated as the correlation between the 10 year term premium
for country i and the United States using all non-FOMC days for the entire sample period of January 1995 to
June 2019. Bond substitutability is standardized to the interval 0 to 1, representing a range in the correlation
between -1 and 1. The sample consists of 204 FOMC announcements from January 1995 to June 2019. All
term premium changes are calculated in a two day window around FOMC announcements. Standard errors
reported in parentheses are calculated with two-way clustering (along the country and time dimension).

154

Table G8: Response of US holdings of foreign bonds to monetary shocks

mps

mpu

idiﬀ

mps x idiﬀ

mpu x idiﬀ

Constant

Advanced
-0.034*
[0.019]
-0.017
[0.033]
-0.003
[0.014]
0.000
[0.010]
0.004
[0.008]
0.129***
[0.024]

Observations
R-squared

3,528
0.052

Emerging

0.082
[0.104]
-0.225**
[0.066]
-0.023
[0.016]
0.014
[0.016]
-0.023***
[0.005]
-0.033
[0.091]

929
0.061

The table shows the response of changes in US holdings of foreign bonds to a monetary policy surprise (mps),
monetary policy uncertainty (mpu) shock, and their interaction with the interest rate diﬀerential between
the 3 month rate in foreign countries relative to the US (idiﬀ). US holdings of foreign bonds are from the
monthly TIC data. Country ﬁxed eﬀects and year dummies are included in the speciﬁcation. The sample
runs from January 1995 to December 2018 for a total of 187 FOMC meetings, which excludes the ﬁnancial
crisis period from December 2007 to June 2009. Standard errors reported in parentheses are calculated with
two-way clustering (along the country and time dimension).

155

Table G9: Understanding the cross-country heterogeneity of asset price responses: Emerging
countries

2 year yield

10 year yield

mpu

FinDepth*mpu

KAopen*mpu

FXRegime*mpu

IRDiﬀ3mChg*mpu

TradeOpen*mpu

Observations
R-squared

0.155***
[0.030]
0.010
[0.040]
0.120**
[0.039]
0.077
[0.128]
-0.037
[0.039]
-0.059
[0.034]

1,056
0.0784

0.168***
[0.045]
0.022
[0.056]
0.088**
[0.033]
0.221
[0.159]
-0.026
[0.031]
-0.085*
[0.040]

1,056
0.0739

The table shows the response of 2 and 10 year government bond yields to a monetary policy uncertainty
(mpu) shock and the interactions with measures for ﬁnancial depth (FinDepth), capital account openness
(KAopen), exchange rate regime (FXRegime), the change in the 3 month interest rate diﬀerential with the
US on an FOMC day (IRiﬀ3mChg), and trade openness (TradeOpen). The monetary policy surprise (mps),
its interactions with the country measures and a constant are included in the regressions, the coeﬃcients
are left out for space considerations. These observables are orthogonalized recursively as in Iacoviello &
Navarro (2019). See Section 3.4.4 for details on the speciﬁcation and variable creation. The sample consists
of 204 FOMC announcements from January 1995 to June 2019. All changes are calculated in a two day
window around FOMC announcements. Standard errors reported in parentheses are calculated with two-way
clustering (along the country and time dimension).

156

Table G10: Response of international equity indices to monetary shocks

Advanced countries

Full sample

Pre-crisis

Post-crisis

mps

mpu

Constant

Observations
R-squared

-0.081
[0.064]

0.098*
[0.048]

5,129
0.005

-0.042
[0.068]
-0.084
[0.062]
0.057
[0.059]

5,129
0.011

-0.031
[0.085]

0.217***
[0.067]

-0.047
[0.093]
0.044
[0.078]
0.237***
[0.076]

2,441
0.001

2,441
0.002

Emerging countries

-0.157*
[0.085]

-0.010
[0.066]

2,688
0.019

-0.052
[0.078]
-0.191**
[0.085]
-0.106
[0.076]

2,688
0.047

Full sample

Pre-crisis

Post-crisis

mps

mpu

Constant

-0.175***
[0.056]

0.128***
[0.041]

-0.125**
[0.048]
-0.110*
[0.061]
0.074
[0.048]

-0.125**
[0.056]

0.208***
[0.050]

-0.123**
[0.053]
-0.005
[0.061]
0.206***
[0.060]

Observations
R-squared

3,102
0.025

3,102
0.034

1,592
0.013

1,592
0.013

-0.270**
[0.098]

0.043
[0.062]

1,510
0.055

-0.144*
[0.071]
-0.228**
[0.101]
-0.071
[0.065]

1,510
0.094

The table shows the response of returns on international equity indices to a monetary policy surprise (mps)
and monetary policy uncertainty (mpu) shock. The full sample consists of 204 FOMC announcements
from January 1995 to June 2019, the pre-crisis sample has 108 announcements from January 1995 to
November 2007 and the post-crisis sample has 96 announcements from December 2007 to June 2019.
Equity returns have been normalized to have unit standard deviation. All changes are calculated in a two
day window around FOMC announcements. Standard errors reported in parentheses are calculated with
two-way clustering (along the country and time dimension).

157

Table G11: Response of international bond yields accounting for zero lower bound

Advanced countries

Emerging countries

mps

mpu

mps*ZLB

mpu*ZLB

ZLB

Constant

2 Year Yield
0.386***
[0.064]
0.065
[0.067]
-0.431***
[0.134]
0.252**
[0.100]
-0.006
[0.085]
-0.033
[0.053]

Observations
R-squared

4,154
0.158

10 Year Yield

2 Year Yield

10 Year Yield

0.224***
[0.072]
0.226***
[0.079]
-0.076
[0.232]
0.113
[0.151]
-0.061
[0.133]
0.031
[0.058]

4,154
0.159

0.156**
[0.047]
0.165**
[0.062]
0.031
[0.100]
-0.022
[0.098]
0.073
[0.077]
-0.027
[0.037]

1,270
0.066

0.110*
[0.056]
0.113***
[0.032]
0.187
[0.172]
0.051
[0.110]
0.113
[0.090]
-0.083***
[0.021]

1,270
0.067

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps)
and monetary policy uncertainty (mpu) shock. All variables have been normalized to have unit standard
deviation. The sample consists of 204 FOMC announcements from January 1995 to June 2019. All changes
are calculated in a two day window around FOMC announcements. The zero lower bound (ZLB) dummy
takes on a value of one from December 2008 to December 2015. Standard errors reported in parentheses are
calculated with two-way clustering (along the country and time dimension).

158

Table G12: Response of forecast dispersion to monetary shocks

Panel (a)

mps

mpu

Constant

Observations
R-squared

Panel (b)

mps

mpu

Constant

Observations
R-squared

SPF forecast

GDP Deﬂator

GDP

0.32
[0.49]
0.52
[0.84]
0.07
[0.04]

97
0.023

0.37
[0.23]
0.75**
[0.30]
0.02
[0.02]

97
0.113

SEP forecast

PCE inﬂation Unemployment

0.01
[0.09]
0.12**
[0.05]
0.04
[0.05]

39
0.119

-0.02
[0.07]
0.09*
[0.05]
0.01
[0.04]

39
0.097

The table shows the results of regressing the changes in the cross-sectional dispersion of SPF forecasts
(GDP and GDP deﬂator) and SEP forecasts (PCE inﬂation and unemployment rate) on the monetary policy
shock measures. The SPF regression (Panel (a)) is at a quarterly frequency with the monetary policy shock
series summed for a given quarter. The SEP regression (Panel (b)) is at the FOMC meeting frequency.
Heteroskedasticity-robust standard errors are reported in parentheses.

159

Table G13: Data coverage for sample countries

Emerging Economy?

Exchange Rate

Equity Index

Yields

Argentina
Australia
Austria
Belgium
Canada
Chile
China
Czech Republic
Denmark
Estonia
Finland
France
Germany
Hong Kong
Hungary
Iceland
India
Indonesia
Ireland
Israel
Italy
Japan
Korea, Republic of
Malaysia
Mauritius
Netherlands
New Zealand
Norway
Pakistan
Peru
Philippines
Poland
Portugal
Russian Federation
Singapore
Slovakia
Slovenia
South Africa
Spain
Sweden
Switzerland
Thailand
Turkey
United Kingdom
United States

Y

Y
Y

Y

Y
Y

Y
Y

Y
Y
Y
Y

Y

Y

Y
Y

8/23/2010

2/3/1999

8/23/2010
2/1/1995

6/26/2002
2/1/1995
2/3/1999
2/3/1999

2/1/1995
5/9/2007

X
X
X
X
X
X

X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X

X
X
X
X
X
X
X

5/20/1997

X
X

X
X
X

X

X

X
X
X
X

X
X
X
X

X
X
X
X

X

X
X
X

2/4/1998

2/3/1999
1/3/2001
1/31/1996

2/3/1999
9/30/1997
10/5/1999
2/3/1999
6/26/2002
7/6/1995
2/3/1999
1/31/1996

8/22/1995

2/1/1995
5/19/1998
5/19/1998
2/1/1995

9/12/2003
7/1/1998
2/1/1995

7/1/1998
5/19/1998

X

7/1/1998
7/1/1998

11/17/1998
3/18/2003
5/19/1998

2/4/1998

X

10/5/1999

5/19/1998
2/1/1995
8/18/1998

7/1/1998
5/19/1998
1/31/2007
7/1/1998

2/1/1995
5/19/1998
2/1/1995
2/1/1995

2/1/1995
2/1/1995

This table displays the availability for each country’s data series. The full sample period covers all FOMC announce-
ments between January 1995 and June 2019, excluding the announcements on Sep. 17, 2001, Oct. 8, 2008, May 22,
2013 and May 2, 2018. Cells with an X indicate a start date of Jan. 1, 1995, cells with a date indicate the ﬁrst date
with available data, and blank cells indicate the data series was not available.

160

Table G14: Response of international bond yields to monetary shocks with QE dummy

Advanced countries

Emerging countries

mps

mpu

mps*QE

mpu*QE

QE

Constant

2 Year Yield
0.365***
[0.063]
0.091*
[0.053]
-0.439
[0.310]
0.292
[0.222]
0.016
[0.202]
-0.048
[0.041]

Observations
R-squared

4,154
0.150

10 Year Yield

2 Year Yield

10 Year Yield

0.205***
[0.070]
0.233***
[0.061]
0.203
[0.500]
0.063
[0.422]
0.054
[0.421]
0.007
[0.048]

4,154
0.160

0.140**
[0.044]
0.117***
[0.033]
0.150
[0.151]
0.153
[0.177]
0.268
[0.180]
-0.025
[0.028]

1,270
0.075

0.098
[0.055]
0.099**
[0.029]
0.235
[0.206]
0.305*
[0.159]
0.557***
[0.155]
-0.075**
[0.022]

1,270
0.096

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps),
monetary policy uncertainty (mpu) shock, and interactions with a quantitative easing dummy. The QE
dummy is 1 for the dates listed in Fawley et al. (2013) and 0 otherwise. All variables have been normalized
to have unit standard deviation. The sample consists of 204 FOMC announcements from January 1995 to
June 2019. All changes are calculated in a two day window around FOMC announcements. Standard errors
reported in parentheses are calculated with two-way clustering (along the country and time dimension).

161

Table G15: Response of international bond yields to monetary shocks (Pre-crisis sample)

Advanced countries

2 year yield
(1)
(2)

0.502***
[0.074]

-0.026
[0.055]

0.452***
[0.073]
0.131**
[0.062]
0.036
[0.060]

mps

mpu

Constant

Observations
R-squared

2,042
0.216

2,042
0.231

10 year yield
(1)
(2)

0.373***
[0.099]

-0.009
[0.070]

2,042
0.119

0.289***
[0.094]
0.220**
[0.086]
0.094
[0.074]

2,042
0.160

Emerging countries

2 year yield
(1)
(2)

mps

mpu

Constant

0.181**
[0.067]

-0.062
[0.045]

Observations
R-squared

502
0.027

0.160**
[0.065]
0.053
[0.063]
-0.038
[0.054]

502
0.029

10 year yield
(2)
(1)
0.118
0.113
[0.090]
[0.082]
0.014
[0.047]
-0.089**
[0.026]

-0.095***
[0.021]

502
0.011

502
0.012

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps)
and monetary policy uncertainty (mpu) shock. All variables have been normalized to have unit standard
deviation. Column (1) has only mps as a regressor, while column 2 adds mpu to this speciﬁcation. The
sample consists of 108 FOMC announcements from January 1995 to November 2007. All changes are
calculated in a two day window around FOMC announcements. Standard errors reported in parentheses are
calculated with two-way clustering (along the country and time dimension).

162

Table G16: Response of international asset prices to monetary shocks (Post-crisis sample)

Advanced countries

2 year yield
(1)

(2)

0.340***
[0.090]

-0.191***
[0.051]

0.275***
[0.089]
0.117
[0.080]
-0.132**
[0.059]

mps

mpu

Constant

0.313**
[0.127]

10 year yield
(2)
(1)
0.128
[0.079]
0.337***
[0.088]
-0.051
[0.081]

-0.222***
[0.071]

Observations
R-squared

2,112
0.089

2,112
0.100

2,112
0.076

2,112
0.162

Emerging countries

2 year yield
(1)

(2)

10 year yield
(1)
(2)

0.325***
[0.081]

-0.103*
[0.046]

0.178***
[0.041]
0.268***
[0.059]
0.032
[0.047]

0.327***
[0.092]

-0.123**
[0.049]

0.171**
[0.055]
0.284***
[0.072]
0.021
[0.056]

mps

mpu

Constant

Observations
R-squared

768
0.082

768
0.137

768
0.083

768
0.145

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps)
and monetary policy uncertainty (mpu) shock. All variables have been normalized to have unit standard
deviation. Column (1) has only mps as a regressor, while column 2 adds mpu to this speciﬁcation. The
sample consists of 96 FOMC announcements from December 2007 to June 2019. All changes are calculated
in a two day window around FOMC announcements. Standard errors reported in parentheses are calculated
with two-way clustering (along the country and time dimension

163

Table G17: Response of US bond term premia to monetary policy shocks

JSZ

ACM

KW

2 year
0.225
[0.125]
0.256*
[0.139]
0.323***
[0.099]

10 year
-0.124
[0.099]
0.393***
[0.113]
0.344***
[0.091]

2 year
0.119
[0.094]
0.373***
[0.085]
0.110
[0.073]

10 year
-0.009
[0.080]
0.458***
[0.079]
0.140**
[0.070]

2 year
0.571***
[0.122]
0.246***
[0.097]
0.155***
[0.076]

10 year
0.310***
[0.088]
0.395***
[0.088]
0.164***
[0.075]

mps

mpu

Constant

Observations
R-squared

204
0.151

204
0.130

204
0.185

204
0.207

204
0.437

204
0.329

The table shows the response of the expected component (ec) and term premium (tp) of 2 and 10 year
government bond yields to a monetary policy uncertainty (mpu) shock. Yields are decomposed into the
expected component and term premium using the methodology of Joslin et al. (2011) (JSW), Adrian et al.
(2013) (ACM) and Kim & Wright (2005) (KW). All variables have been normalized to have unit standard
deviation. Panel (b) adds the US 10 year yield term premium to the speciﬁcation. The sample consists of 204
FOMC announcements from January 1995 to June 2019. All changes are calculated in a two day window
around FOMC announcements. Heteroskedasticity-robust standard errors are reported in parentheses.

164

Table G18: Understanding the cross-country heterogeneity of asset price responses: Advanced
countries

2 year yield

10 year yield

mpu

IRDiﬀ3mChg*mpu

FinDepth*mpu

FXRegime*mpu

TradeOpen*mpu

Observations
R-squared

0.111***
[0.035]
-0.075***
[0.020]
-0.039
[0.042]
0.039
[0.036]
-0.012
[0.022]

3,391
0.170

0.251***
[0.053]
-0.026
[0.027]
0.007
[0.048]
0.001
[0.052]
0.018
[0.024]

3,391
0.179

The table shows the response of 2 and 10 year government bond yields to a monetary policy uncertainty
(mpu) shock and the interactions with measures for ﬁnancial depth (FinDepth), capital account openness
(KAopen), exchange rate regime (FXRegime), the change in the 3 month interest rate diﬀerential with the
US on an FOMC day (IRiﬀ3mChg), and trade openness (TradeOpen). These observables are orthogonalized
recursively as in Iacoviello & Navarro (2019). See Section 3.4.4 for details on the speciﬁcation and variable
creation. The sample consists of 204 FOMC announcements from January 1995 to June 2019. All changes
are calculated in a two day window around FOMC announcements. Standard errors reported in parentheses
are calculated with two-way clustering (along the country and time dimension).

165

Table G19: Understanding the cross-country heterogeneity of asset price responses: Emerging
countries with dollar debt exposure

2 year yield

10 year yield

mpu

TradeOpen*mpu

KAopen*mpu

FXRegime*mpu

DollarDebt*mpu

IRDiﬀ3mChg*mpu

Observations
R-squared

0.124***
[0.030]
0.007
[0.041]
0.193***
[0.047]
-0.068
[0.138]
0.094
[0.079]
-0.046
[0.039]

808
0.0660

0.134**
[0.043]
-0.013
[0.033]
0.146*
[0.069]
0.134
[0.168]
0.154*
[0.064]
-0.034
[0.029]

808
0.0559

The table shows the response of 2 and 10 year government bond yields to a monetary policy uncertainty
(mpu) shock and the interactions with measures for trade openness (TradeOpen), capital account openness
(KAopen), exchange rate regime (FXRegime), dollar debt exposure (DollarDebt), and the change in the
3 month interest rate diﬀerential with the US on an FOMC day (IRiﬀ3mChg). These observables are
orthogonalized recursively as in Iacoviello & Navarro (2019). See Section 3.4.4 for details on the speciﬁcation
and variable creation. The sample consists of 204 FOMC announcements from January 1995 to June 2019.
All changes are calculated in a two day window around FOMC announcements. Standard errors reported in
parentheses are calculated with two-way clustering (along the country and time dimension).

166

Table G20: Response of international bond yields to monetary shocks (1-day window)

Advanced countries

2 year yield
(1)
(2)

10 year yield
(1)
(2)

mps

mpu

Constant

0.340***
[0.075]

-0.086**
[0.038]

0.263***
[0.070]
0.192***
[0.055]
0.045
[0.046]

Observations
R-squared

4,154
0.101

4,154
0.132

0.266**
[0.096]

-0.084*
[0.048]

4,154
0.062

0.164*
[0.084]
0.257***
[0.049]
0.092
[0.059]

4,154
0.117

Emerging countries

2 year yield
(1)
(2)

0.196***
[0.046]

-0.086**
[0.036]

0.131***
[0.036]
0.161**
[0.063]
0.021
[0.043]

mps

mpu

Constant

0.139*
[0.066]

10 year yield
(2)
(1)
0.092
[0.059]
0.117*
[0.053]
0.003
[0.038]

-0.075*
[0.033]

Observations
R-squared

1,270
0.034

1,270
0.055

1,270
0.017

1,270
0.028

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps)
and monetary policy uncertainty (mpu) shock. All variables have been normalized to have unit standard
deviation. Column (1) has only mps as a regressor, while column 2 adds mpu to this speciﬁcation. The
sample consists of 204 FOMC announcements from January 1995 to June 2019. All changes are calculated
in a one day window around FOMC announcements. Standard errors reported in parentheses are calculated
with two-way clustering (along the country and time dimension).

167

Table G21: Response of international bond yields to monetary shocks, controlling for LIBOR-OIS
spread

Advanced countries

2 year yield
(2)
(1)

0.382***
[0.079]

-0.002
[0.021]
-0.099**
[0.043]

0.301***
[0.077]
0.139**
[0.056]
0.002
[0.017]
-0.032
[0.047]

10 year yield
(2)
(1)

0.341***
[0.115]

-0.014
[0.010]
-0.123**
[0.058]

0.162*
[0.086]
0.305***
[0.072]
-0.005
[0.008]
0.023
[0.064]

mps

mpu

LIBOR-OIS Spread

Constant

Observations
R-squared

3,212
0.115

3,212
0.132

3,212
0.099

3,212
0.173

Emerging countries

2 year yield
(1)
(2)

mps

mpu

LIBOR-OIS Spread

Constant

Observations
R-squared

0.230***
[0.054]

-0.006
[0.006]
-0.071*
[0.031]

1,103
0.051

0.137**
[0.042]
0.164***
[0.041]
-0.001
[0.004]
0.008
[0.035]

0.195**
[0.062]

10 year yield
(1)
(2)
0.104
[0.057]
0.159**
[0.046]
-0.033***
[0.007]
-0.019
[0.037]

-0.038***
[0.006]
-0.096**
[0.030]

1,103
0.076

1,103
0.089

1,103
0.114

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps)
and monetary policy uncertainty (mpu) shock while controlling for the LIBOR-OIS spread. The sample
consists of 146 FOMC announcements from December 2001 (when LIBOR-OIS data becomes available) to
June 2019. All changes are calculated in a two day window around FOMC announcements. Standard errors
reported in parentheses are calculated with two-way clustering (along the country and time dimension).

168

Table G22: Response of international bond yields to monetary shocks, 2 year yield as mps

Advanced countries

2 year yield
(2)
(1)

10 year yield
(2)
(1)

mps (2yr)

mpu

Constant

0.389***
[0.051]

-0.065*
[0.036]

0.345***
[0.051]
0.125**
[0.051]
-0.008
[0.038]

Observations
R-squared

4,154
0.146

4,154
0.160

0.399***
[0.053]

-0.076
[0.045]

4,154
0.154

0.313***
[0.050]
0.240***
[0.053]
0.034
[0.048]

4,154
0.204

Emerging countries

2 year yield
(1)
(2)

10 year yield
(1)
(2)

0.170**
[0.054]

-0.061*
[0.031]

0.101*
[0.051]
0.188***
[0.040]
0.024
[0.030]

0.156**
[0.052]

-0.092**
[0.029]

0.084*
[0.043]
0.194***
[0.051]
-0.005
[0.030]

mps (2yr)

mpu

Constant

Observations
R-squared

1,270
0.025

1,270
0.056

1,270
0.021

1,270
0.055

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps)
(measured as the change in the 2 year Treasury yield) and monetary policy uncertainty (mpu) shock. All
variables have been normalized to have unit standard deviation. Column (1) has only mps as a regressor,
while column 2 adds mpu to this speciﬁcation. The sample consists of 204 FOMC announcements from
January 1995 to June 2019. All changes are calculated in a two day window around FOMC announcements.
Standard errors reported in parentheses are calculated with two-way clustering (along the country and time
dimension).

169

Table G23: Response of international bond yields to monetary shocks, 10 year yield as mps

Advanced countries

2 year yield
(2)
(1)

10 year yield
(2)
(1)

mps (10 yr)

0.293***
[0.063]

mpu

Constant

-0.080*
[0.040]

0.230***
[0.067]
0.121*
[0.059]
-0.026
[0.044]

Observations
R-squared

4,154
0.088

4,154
0.099

0.527***
[0.056]

-0.072*
[0.038]

4,154
0.287

0.479***
[0.062]
0.093*
[0.050]
-0.030
[0.042]

4,154
0.293

Emerging countries

2 year yield
(1)
(2)

10 year yield
(1)
(2)

mps (10 yr)

0.174***
[0.044]

mpu

Constant

-0.064*
[0.031]

0.087*
[0.045]
0.173***
[0.044]
0.013
[0.033]

0.215***
[0.039]

-0.092***
[0.026]

0.142***
[0.036]
0.145**
[0.056]
-0.027
[0.031]

Observations
R-squared

1,270
0.032

1,270
0.054

1,270
0.050

1,270
0.065

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps)
(measured as the change in the 10 year Treasury yield) and monetary policy uncertainty (mpu) shock. All
variables have been normalized to have unit standard deviation. Column (1) has only mps as a regressor,
while column 2 adds mpu to this speciﬁcation. The sample consists of 204 FOMC announcements from
January 1995 to June 2019. All changes are calculated in a two day window around FOMC announcements.
Standard errors reported in parentheses are calculated with two-way clustering (along the country and time
dimension).

170

Table G24: Response of expected component and term premium component of international bond
yields to monetary policy uncertainty (pre- and post-crisis samples)

Jan-1995 to Nov-2007

Panel (a)

mpu

Advanced countries
10y ec
-0.011
[0.057]

2y tp
0.147**
[0.067]

10y tp
0.206**
[0.084]

2y ec
-0.013
[0.060]

Panel (b)

mpu

US 10y tp

Advanced countries
10y ec
0.024
[0.055]
-0.101*
[0.052]

2y tp
0.051
[0.046]
0.276***
[0.053]

10y tp
0.052
[0.061]
0.439***
[0.063]

2y ec
0.026
[0.050]
-0.111**
[0.046]

Emerging countries
2y tp
0.049
[0.046]

10y ec
0.017
[0.038]

Emerging countries
2y tp
0.048
[0.052]
0.005
[0.042]

10y ec
0.019
[0.042]
-0.004
[0.024]

2y ec
0.020
[0.065]

2y ec
0.025
[0.071]
-0.017
[0.032]

Panel (c)

mpu

Panel (d)

mpu

US 10y tp

2y ec
-0.012
[0.077]

2y ec
0.052
[0.079]
-0.139*
[0.069]

Dec-2007 to Jun-2019

Advanced countries
10y ec
0.010
[0.090]

2y tp
0.141*
[0.080]

10y tp
0.297***
[0.099]

2y ec

0.192***
[0.028]

Emerging countries
2y tp
0.084
[0.051]

10y ec
0.246***
[0.056]

Advanced countries
10y ec
0.051
[0.098]
-0.089
[0.077]

2y tp
0.047
[0.060]
0.205***
[0.064]

10y tp
0.103
[0.085]
0.422***
[0.055]

Emerging countries
2y tp
0.034
[0.041]
0.109**
[0.037]

10y ec
0.244***
[0.053]
0.003
[0.049]

2y ec

0.214***
[0.032]
-0.049
[0.066]

10y tp
0.007
[0.030]

10y tp
0.006
[0.032]
0.001
[0.042]

10y tp
0.094
[0.060]

10y tp
0.055
[0.076]
0.084
[0.070]

The table shows the response of the expected component (ec) and term premium (tp) of 2 and 10 year
government bond yields to a monetary policy uncertainty (mpu) shock. Yields are decomposed into the
expected component and term premium using the methodology of Joslin et al. (2011). All variables have
been normalized to have unit standard deviation. Panels (a) and (b) show results for the pre-crisis sample,
consisting of 108 FOMC announcements from January 1995 to November 2007. Panels (c) and (d) show
results for the post-crisis sample, consisting of 96 FOMC announcements from December 2007 to June
2019. Panels (a) and (c) report the eﬀects of mpu only. Panels (b) and (d) add the US 10 year yield term
premium to the speciﬁcation. All changes are calculated in a two day window around FOMC announcements.
Standard errors reported in parentheses are calculated with two-way clustering (along the country and time
dimension).

171

Table G25: Response of term premium component of international bond yields to mpu shocks
(bond substitutability interaction)

10 year term premium

All Countries Advanced Emerging

mpu

mpu x bond sub.

Observations
R-squared

-0.16**
(0.069)
0.50***
(0.120)

5,410
0.044

-0.08
(0.114)
0.42**
(0.156)

4,140
0.062

-0.10
(0.102)
0.26
(0.197)

1,270
0.008

The table shows the response of 10 year government bond yield term premia to a monetary policy uncertainty
(mpu) shock and the interaction with a measure of bond substitutability with the United States. The term
premium is calculated using the methodology of Joslin et al. (2011). Bond substitutability is calculated as
the correlation between the 10 year term premium for country i and the United States using all non-FOMC
days between January 1995 and the FOMC day on day t . Bond substitutability is standardized to the
interval 0 to 1, representing a range in the correlation between -1 and 1. The sample consists of 204 FOMC
announcements from January 1995 to June 2019. All term premium changes are calculated in a two day
window around FOMC announcements. Standard errors reported in parentheses are calculated with two-way
clustering (along the country and time dimension).

172

Table G26: Response of US holdings of foreign bonds to monetary shocks (robustness)

Advanced

Emerging

mps

mpu

idiﬀ

mps x idiﬀ

mpu x idiﬀ

Constant

-

-

-0.018
[0.015]
-0.001
[0.009]
0.005
[0.006]
0.135***
[0.002]

Observations
R-squared

3,528
0.149

-

-

-0.027
[0.019]
0.013
[0.017]
-0.028**
[0.011]
0.054
[0.069]

903
0.290

The table shows the response of changes in US holdings of foreign bonds to the interaction between a monetary
policy surprise (mps) and monetary policy uncertainty (mpu) shock with the interest rate diﬀerential between
the 3 month rate in foreign countries relative to the US (idiﬀ). US holdings of foreign bonds are from the
monthly TIC data. Country and time ﬁxed eﬀects are included in the speciﬁcation. The sample runs from
January 1995 to December 2018 for a total of 187 FOMC meetings, which excludes the ﬁnancial crisis period
from December 2007 to June 2009. Standard errors reported in parentheses are calculated with two-way
clustering (along the country and time dimension).

173

Table G27: Response of international bond yields to monetary shocks, controlling for target and
path factor

Advanced countries

2 year yield
(1)

(2)

10 year yield
(2)
(1)

0.429***
[0.058]
0.254**
[0.119]

-0.103***
[0.035]

0.386***
[0.058]
0.212
[0.138]
0.086
[0.052]
-0.061
[0.041]

0.369***
[0.066]
0.543***
[0.171]

-0.114**
[0.046]

0.260***
[0.057]
0.435***
[0.145]
0.221***
[0.061]
-0.006
[0.051]

Target

Path

mpu

Constant

Observations
R-squared

4,154
0.147

4,154
0.153

4,154
0.146

4,154
0.184

Emerging countries

2 year yield
(1)

(2)

10 year yield
(1)
(2)

0.247***
[0.054]
0.031
[0.097]

-0.077**
[0.030]

0.159**
[0.048]
-0.045
[0.086]
0.163***
[0.034]
0.001
[0.034]

0.224**
[0.071]
0.048
[0.091]

-0.107***
[0.028]

0.129*
[0.058]
-0.034
[0.074]
0.175***
[0.043]
-0.023
[0.030]

Target

Path

mpu

Constant

Observations
R-squared

1,270
0.044

1,270
0.065

1,270
0.036

1,270
0.060

The table shows the response of 2 and 10 year government bond yields to a target factor shock, path factor
shock, and monetary policy uncertainty (mpu) shock. All variables have been normalized to have unit
standard deviation. Column (1) includes the target and path factors as regressors, while column 2 adds mpu
to the speciﬁcation. The sample consists of 204 FOMC announcements from January 1995 to June 2019.
All changes are calculated in a two day window around FOMC announcements. Standard errors reported in
parentheses are calculated with two-way clustering (along the country and time dimension).

174

Table G28: Response of international bond yields to monetary shocks (only scheduled FOMC
meetings)

Advanced countries

2 year yield
(1)

(2)

10 year yield
(1)
(2)

0.435***
[0.075]

-0.109***
[0.036]

0.362***
[0.078]
0.115**
[0.051]
-0.049
[0.043]

0.440***
[0.068]

-0.133**
[0.048]

0.305***
[0.070]
0.213***
[0.063]
-0.021
[0.055]

mps

mpu

Constant

Observations
R-squared

3,956
0.128

3,956
0.138

3,956
0.128

3,956
0.161

Emerging countries

2 year yield
(1)

(2)

mps

mpu

Constant

Observations
R-squared

0.220***
[0.049]

-0.055
[0.031]

1,209
0.033

0.120*
[0.052]
0.145***
[0.038]
0.020
[0.039]

0.216**
[0.080]

10 year yield
(1)
(2)
0.112
[0.072]
0.152**
[0.051]
-0.019
[0.036]

-0.097**
[0.030]

1,209
0.049

1,209
0.032

1,209
0.050

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps)
and monetary policy uncertainty (mpu) shock. All variables have been normalized to have unit standard
deviation. Column (1) has only mps as a regressor, while column 2 adds mpu to this speciﬁcation. The
sample consists of 195 scheduled FOMC announcements from January 1995 to June 2019. All changes are
calculated in a two day window around FOMC announcements. Standard errors reported in parentheses are
calculated with two-way clustering (along the country and time dimension).

175

Table G29: Response of international bond yields to monetary shocks controlling for information
eﬀect

Advanced countries

2 year yield
(2)
(1)

10 year yield
(2)
(1)

mps

mpu

Constant

0.398***
[0.070]

-0.113**
[0.051]

0.367***
[0.065]
0.079
[0.054]
-0.098*
[0.052]

Observations
R-squared

2,680
0.135

2,680
0.141

0.290***
[0.080]

-0.084
[0.061]

2,680
0.085

0.221***
[0.071]
0.175***
[0.058]
-0.050
[0.059]

2,680
0.119

Emerging countries

2 year yield
(2)
(1)

mps

mpu

Constant

0.151***
[0.037]

-0.078*
[0.040]

Observations
R-squared

739
0.020

0.116**
[0.036]
0.085**
[0.035]
-0.062
[0.041]

739
0.027

0.147*
[0.073]

10 year yield
(2)
(1)
0.103
[0.062]
0.104**
[0.043]
-0.067*
[0.034]

-0.086*
[0.037]

739
0.020

739
0.031

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps)
and monetary policy uncertainty (mpu) shock. mps and mpu have been purged of information eﬀects in
this speciﬁcation. Each monetary shock is regressed on the diﬀerence between Federal Reserve Greenbook
forecasts and private sector Blue Chip forecasts of CPI, GDP, and the unemployment rate. The residual from
these regressions is taken as an information-robust measure of mps and mpu. Column (1) has only mps as a
regressor, while column 2 adds mpu to this speciﬁcation. The sample consists of 204 FOMC announcements
from January 1995 to December 2011. All changes are calculated in a two day window around FOMC
announcements. Standard errors reported in parentheses are calculated with two-way clustering (along the
country and time dimension).

176

Table G30: Response of international bond yields to monetary shocks with country ﬁxed-eﬀects

Advanced countries

2 year yield
(1)

(2)

10 year yield
(1)
(2)

0.410***
[0.065]

-0.103***
[0.033]

0.360***
[0.062]
0.107**
[0.048]
-0.051
[0.040]

0.345***
[0.080]

-0.115**
[0.047]

0.221***
[0.069]
0.264***
[0.060]
0.014
[0.053]

mps

mpu

Constant

Observations
R-squared

4,154
0.140

4,154
0.150

4,154
0.100

4,154
0.157

Emerging countries

2 year yield
(1)

(2)

10 year yield
(1)
(2)

0.239***
[0.051]

-0.077***
[0.020]

0.159***
[0.043]
0.160***
[0.039]
0.000
[0.029]

0.215**
[0.070]

-0.108***
[0.024]

0.129*
[0.057]
0.172***
[0.046]
-0.024
[0.033]

mps

mpu

Constant

Observations
R-squared

1,270
0.048

1,270
0.069

1,270
0.038

1,270
0.062

The table shows the response of 2 and 10 year government bond yields to a monetary policy surprise (mps)
and monetary policy uncertainty (mpu) shock, with country ﬁxed eﬀects included in the speciﬁcation. All
variables have been normalized to have unit standard deviation. Column (1) has only mps as a regressor,
while column 2 adds mpu to this speciﬁcation. The sample consists of 204 FOMC announcements from
January 1995 to June 2019. All changes are calculated in a two day window around FOMC announcements.
Standard errors reported in parentheses are calculated with two-way clustering (along the country and time
dimension).

177

APPENDIX H

FIGURES FOR CHAPTER 3

Figure H1: Monetary policy uncertainty changes (mpu) on FOMC meeting days

The ﬁgure shows the two-day change in the standard deviation of the 1 year ahead rate on FOMC
meeting days (our baseline mpu measure). The measure has been normalized to have unit standard
deviation. The labeled dates are the three largest declines and three largest increases in mpu.

178

Figure H2: 10 Year Yield Response on Prominent Monetary Policy Uncertainty Dates

(a) Advanced: MPU Increase Dates

(b) Advanced: MPU Decrease Dates

(c) Emerging: MPU Increase Dates

(d) Emerging: MPU Decrease Dates

The ﬁgure shows the average total change in 10 year bond yields on ten FOMC dates with the
largest increase or decrease in the monetary policy uncertainty (mpu) shock, along with the average
predicted component due to mpu and the average predicted component due to the monetary policy
surprise (mps). The average predicted components are based on the coeﬃcients estimated in
Equation 3.3.

179

Figure H3: Response of 3 month yield to monetary policy uncertainty

The ﬁgure shows the dynamic response of 3 month government bond yields to a monetary policy
uncertainty (mpu) shock over an 18 month horizon. The change in 3 month yields has been
normalized to have unit standard deviation. The sample consists of 204 FOMC announcements
from January 1995 to June 2019. All changes are calculated in a two day window around FOMC
announcements. 68% conﬁdence bands are constructed from Driscoll and Kraay standard errors.

180

−.2−.10.1.2.3.40369121518MonthsAdvanced−.2−.10.1.2.3.40369121518MonthsEmergingFigure H4: Correlation between change in monetary policy uncertainty (mpu) and monetary policy
surprise (mps)

The ﬁgure plots the monetary policy uncertainty (mpu) shock against the monetary policy surprise
(mps). Both measures are calculated in a two day window around FOMC announcements. The
sample consists of 204 FOMC announcements from January 1995 to June 2019. The diagonal line
shows the ﬁt from the regression of mpu on mps.

181

-5-4-3-2-101234-5-4-3-2-10123Figure H5: Standard deviation of international asset prices

The ﬁgure plots the standard deviation of the 2-day change in the 2 and 10 year bond yields and
stock market return by country.

182

0.1.2.3.42-year yield std. dev.ausautbelcanchnczednkespﬁnfragerhkghunidnindireitajapmysnldnornzlpolprtrussafsgpsweswiukAdvancedEME.511.522.533.54Equity returns std. dev.argausautbelcanchlchnczednkespestﬁnfragerhkghunidnindireislisritajapkormusmysnldnornzlpakperphlpolprtrussafsgpsvksvnsweswithaturuk0.1.2.3.410-year yield std. dev.ausautbelcanchnczednkespﬁnfragerhkghunidnindireitajapmysnldnornzlpolprtrussafsgpsweswiukFigure H6: 10 Year Yield Response on Prominent Monetary Policy Uncertainty Dates

(a) Advanced: MPU Increase Dates

(b) Advanced: MPU Decrease Dates

(c) Emerging: MPU Increase Dates

(d) Emerging: MPU Decrease Dates

The ﬁgure shows the change in 10 year yields, the change attributable to the monetary policy
surprise (mps) and the change attributable to the monetary policy uncertainty (mpu) shock on
FOMC dates with prominent changes in mpu. Panel (a) displays the reaction of advanced yields
for dates with large mpu increases, panel (b) displays the reaction of advanced yields for dates with
large mpu decreases, panel (c) displays the reaction of emerging yields for dates with large mpu
increases, and panel (d) displays the reaction of emerging yields for dates with large decreases in
mpu.

183

Figure H7: Persistence of international asset price response to mpu

The ﬁgure shows the dynamic response of 2 year government bond yields, 10 year month government
bond yields, and equity returns to a monetary policy uncertainty (mpu) shock over a 45 day horizon.
The change in asset prices have been normalized to have unit standard deviation. The sample
consists of 204 FOMC announcements from January 1995 to June 2019. mpu is calculated over a
two day window around FOMC announcements. 95% and 68% conﬁdence bands are constructed
from Driscoll and Kraay standard errors.

184

-.50.510153045DaysAdvanced: 2 Year Response-.50.511.50153045DaysEmerging: 2 Year Response-.50.510153045DaysAdvanced: 10 Year Response-.50.510153045DaysEmerging: 10 Year Response-1-.50.50153045DaysAdvanced: Equity Response-1-.50.50153045DaysEmerging: Equity ResponseFigure H8: Heterogeneity in response to mpu across countries

(a) 2 year yields

(b) 10 year yields

The ﬁgure shows country speciﬁc responses to a monetary policy uncertainty (mpu) shock. Panel (a)
shows the response of 2 year yields and panel (b) shows the response of 10 year yields. Coeﬃcients
are estimated for the full sample period available for each country. 90% conﬁdence intervals are
reported. The dashed-with-3-dots line is the pooled OLS estimate. The long-dashed line is the
emerging OLS estimate. The short-dashed line is the advanced OLS estimate.

Figure H9: Robustness of capital account openness interaction with mpu

(a) 2 year yields

(b) 10 year yields

The ﬁgure shows the KAopen*mpu coeﬃcient from Table G9 for the 24 unique variable orderings
with ﬁnancial depth appearing ﬁrst in the orthogonalization. The top estimate is the baseline
speciﬁcation. 90% conﬁdence intervals are reported.

185

Figure H10: 2 and 10 Year Yield Response to Monetary Policy Uncertainty. Dropping One Country
at a Time.

(a) Advanced: 2 Year Yield

(b) Advanced: 10 Year Yield

(c) Emerging: 2 Year Yield

(d) Emerging: 10 Year Yield

The ﬁgure shows the response of 2 and 10 year government bond yields to a monetary policy
uncertainty (mpu) shock while dropping one country at a time from the advanced and emerging
country samples, respectively. The sample consists of 204 FOMC announcements from January
1995 to June 2019. All changes are calculated in a two day window around FOMC announcements.
Conﬁdence intervals are constructed with two-way clustered standard errors (along the country and
time dimension).

186

Figure H11: 2 and 10 Year Yield Response to Monetary Policy Uncertainty. Dropping One FOMC
Date at a TIme.

(a) Advanced: 2 Year Yield

(b) Advanced: 10 Year Yield

(c) Emerging: 2 Year Yield

(d) Emerging: 10 Year Yield

The ﬁgure shows the response of 2 and 10 year government bond yields to a monetary policy
uncertainty (mpu) shock while dropping one FOMC date at a time. The sample consists of 204
FOMC announcements from January 1995 to June 2019. All changes are calculated in a two
day window around FOMC announcements. Conﬁdence intervals are constructed with two-way
clustered standard errors (along the country and time dimension).

187

BIBLIOGRAPHY

188

BIBLIOGRAPHY

Adrian, Tobias, Richard K Crump & Emanuel Moench. 2013. Pricing the term structure with linear

regressions. Journal of Financial Economics 110(1). 110–138.

Ahmed, Shaghil & Andrei Zlate. 2014. Capital ﬂows to emerging market economies: A brave new

world? Journal of International Money and Finance 48. 221–248.

Albagli, Elias, Luis Ceballos, Sebastian Claro & Damian Romero. 2019. Channels of us monetary

policy spillovers to international bond markets. Journal of Financial Economics .

Alpanda, Sami & Serdar Kabaca. 2019. International Spillovers of Large-Scale Asset Purchases.

Journal of the European Economic Association Jvy053.

Anaya, Pablo, Michael Hachula & Christian J Oﬀermanns. 2017. Spillovers of us unconventional
monetary policy to emerging markets: The role of capital ﬂows. Journal of International Money
and Finance 73. 275–295.

Bacchetta, Philippe, Kenza Benhima & Céline Poilly. 2019. Corporate cash and employment.

American Economic Journal: Macroeconomics 11(3). 30–66.

Bauer, Michael, Aeimit Lakdawala & Philippe Mueller. 2019. Market-based monetary policy

uncertainty. Federal Reserve Bank of San Francisco Working Paper .

Bauer, Michael D & Christopher J Neely. 2014. International channels of the fed’s unconventional

monetary policy. Journal of International Money and Finance 44. 24–46.

Bauer, Michael D & Glenn D Rudebusch. 2014. The signaling channel for federal reserve bond

purchases. International Journal of Central Banking .

Bauer, Michael D, Glenn D Rudebusch & Jing Cynthia Wu. 2012. Correcting estimation bias in

dynamic term structure models. Journal of Business & Economic Statistics 30(3). 454–467.

Begenau, Juliane & Juliana Salomao. 2019. Firm ﬁnancing over the business cycle. The Review of

Financial Studies 32(4). 1235–1274.

Berger, Allen, Rebecca Demsetz & Philip Strahan. 1999. The consolidation of the ﬁnancial services
industry: Causes, consequences, and implications for the future. Journal of Banking & Finance
23(2-4). 135–194.

Berger, Allen, Nathan Miller, Mitchell Petersen, Raghuram Rajan & Jeremy Stein. 2005. Does
function follow organizational form? Evidence from the lending practices of large and small
banks. Journal of Financial Economics 76(2). 237–269.

Bernanke, Ben. 2015.

Term premi-
ums. https://www.brookings.edu/blog/ben-bernanke/2015/04/13/why-are-interest-rates-so-low-
part-4-term-premiums/ .

so low,

part 4:

Why are

interest

rates

189

Bernanke, Ben, Mark Gertler & Simon Gilchrist. 1999. The ﬁnancial accelerator in a quantitative

business cycle framework. Handbook of Macroeconomics 1. 1341–1393.

Bernanke, Ben S & Mark Gertler. 1995. Inside the black box: the credit channel of monetary policy

transmission. Journal of Economic perspectives 9(4). 27–48.

Bernanke, Ben S & Kenneth N Kuttner. 2005. What explains the stock market’s reaction to federal

reserve policy? The Journal of ﬁnance 60(3). 1221–1257.

Bertaut, Carol C & Ruth Judson. 2014. Estimating us cross-border securities positions: New data

and new methods. FRB International Finance Discussion Paper (1113).

Bertaut, Carol C & Ralph W Tryon. 2007. Monthly estimates of us cross-border securities positions.

FRB International Finance Discussion Paper (910).

Bhattarai, Saroj, Arpita Chatterjee & Woong Yong Park. 2015. Eﬀects of us quantitative easing on

emerging market economies .

Bhattarai, Saroj, Arpita Chatterjee & Woong Yong Park. 2019. Global spillover eﬀects of us

uncertainty. Journal of Monetary Economics .

Bhattarai, Saroj & Christopher Neely. 2016. A survey of the empirical literature on us unconven-

tional monetary policy. Federal Reserve Bank of St. Louis, Research Division.

Boot, Arnoud. 2000. Relationship banking: What do we know? Journal of Financial Intermedia-

tion 9. 7–25.

Bowman, David, Juan M Londono & Horacio Sapriza. 2015. Us unconventional monetary policy
and transmission to emerging market economies. Journal of International Money and Finance
55. 27–59.

Broner, Fernando, Tatiana Didier, Aitor Erce & Sergio L Schmukler. 2013. Gross capital ﬂows:

Dynamics and crises. Journal of Monetary Economics 60. 113–133.

Bundick, Brent, Trenton Herriford & A. Lee Smith. 2017a. Forward guidance, monetary policy

uncertainty, and the term premium. Working Paper, Federal Reserve Bank of City.

Bundick, Brent, Trenton Herriford & Andrew Smith. 2017b. Forward guidance, monetary policy
uncertainty, and the term premium. Monetary Policy Uncertainty, and the Term Premium (July
2017) .

Bustamante, Christian. 2018. Debt overhang, monetary policy, and economic recoveries after large

recessions. Working Paper .

Campbell, Jeﬀrey R, Charles L Evans, Jonas DM Fisher, Alejandro Justiniano, Charles W Calomiris
& Michael Woodford. 2012. Macroeconomic eﬀects of federal reserve forward guidance [with
comments and discussion]. Brookings Papers on Economic Activity 1–80.

Carow, Kenneth, Edward Kane & Rajesh Narayanan. 2003. How have borrowers fared in banking

mega-mergers? NBER Working Paper (w10623).

190

Carrière-Swallow, Yan & Luis Felipe Céspedes. 2013. The impact of uncertainty shocks in emerging

economies. Journal of International Economics 90(2). 316–325.

Chang, Xin, Sudipto Dasgupta, George Wong & Jiaquan Yao. 2014. Cash-ﬂow sensitivities and

the allocation of internal cash ﬂow. The Review of Financial Studies 27(12). 3628–3657.

Chari, Anusha, Karlye Dilts Stedman & Christian Lundblad. 2020. Taper Tantrums: Quantitative
Easing, Its Aftermath, and Emerging Market Capital Flows. The Review of Financial Studies .

Chinn, Menzie D & Hiro Ito. 2006. What matters for ﬁnancial development? capital controls,

institutions, and interactions. Journal of development economics 81(1). 163–192.

Christiano, Lawrence J, Martin Eichenbaum & Charles L Evans. 1999. Monetary policy shocks:

What have we learned and to what end? Handbook of macroeconomics 1. 65–148.

Cloyne, James, Clodomiro Ferreira, Maren Froemel & Paolo Surico. 2018. Investment, ﬁnancial

frictions and the dynamic eﬀects of monetary policy. Tech. rep. Working paper.

Covas, Francisco & Wouter Den Haan. 2011. The cyclical behavior of debt and equity ﬁnance.

American Economic Review 101(2). 877–99.

Covas, Francisco & Wouter Den Haan. 2012. The role of debt and equity ﬁnance over the business

cycle. The Economic Journal 122(565). 1262–1286.

Creal, Drew D & Jing Cynthia Wu. 2017. Monetary policy uncertainty and economic ﬂuctuations.

International Economic Review 58(4). 1317–1354.

Crouzet, Nicolas. 2020. Credit disintermediation and monetary policy Mimeo, Northwestern

University.

Crouzet, Nicolas, Neil Mehrotra et al. 2017. Small and large ﬁrms over the business cycle.

Unpublished working paper .

Curcuru, Stephanie E, Steven B Kamin, Canlin Li & Marius Rodriguez. 2018.

International
spillovers of monetary policy: Conventional policy vs. quantitative easing. FRB International
Finance Discussion Paper (1234).

Daetz, Stine Louise, Marti G Subrahmanyam, Dragon Yongjun Tang & Sarah Qian Wang. 2018. Can
central banks boost corporate investment: Evidence from the ecb liquidity injections. Available
at SSRN 2915800 .

Dahlhaus, Tatjana & Garima Vasishtha. 2014. The impact of us monetary policy normalization on

capital ﬂows to emerging-market economies. Tech. rep. Bank of Canada working paper.

Davis, Steven, John Haltiwanger, Ron Jarmin, Javier Miranda, Christopher Foote & Eva Nagypal.
2006. Volatility and dispersion in business growth rates: Publicly traded versus privately held
ﬁrms. NBER Macroeconomics Annual 21. 107–179.

Dedola, Luca & Francesco Lippi. 2005. The monetary transmission mechanism: evidence from

the industries of ﬁve oecd countries. European Economic Review 49(6). 1543–1569.

191

Delis, Manthos, Sotirios Kokas & Steven Ongena. 2017. Bank market power and ﬁrm performance.

Review of Finance 21(1). 299–326.

Di Patti, Emilia Bonaccorsi & Giorgio Gobbi. 2007. Winners or losers? The eﬀects of banking

consolidation on corporate borrowers. The Journal of Finance 62(2). 669–695.

Diallo, Ibrahima. 2018. HAMILTONFILTER: Stata module to calculate the Hamilton ﬁlter for a

single time series or for a panel dataset .

Diamond, Douglas & Raghuram Rajan. 2001. Liquidity risk, liquidity creation, and ﬁnancial

fragility: A theory of banking. Journal of Political Economy 109(2). 287–327.

Dick, Astrid A. 2006. Nationwide branching and its impact on market structure, quality, and bank

performance. The Journal of Business 79(2). 567–592.

Ehrmann, Michael & Marcel Fratzscher. 2004. Taking stock: Monetary policy transmission to

equity markets. Journal of Money, Credit and Banking 719–737.

Ehrmann, Michael, Marcel Fratzscher & Roberto Rigobon. 2011. Stocks, bonds, money markets and
exchange rates: measuring international ﬁnancial transmission. Journal of Applied Econometrics
26(6). 948–974.

Eisfeldt, Andrea & Tyler Muir. 2016. Aggregate external ﬁnancing and savings waves. Journal of

Monetary Economics 84. 116–133.

Fawley, Brett W, Christopher J Neely et al. 2013. Four stories of quantitative easing. Federal

Reserve Bank of St. Louis Review 95(1). 51–88.

Ferguson, Roger. 2001.

April 19th Remarks by Vice Chairman Roger W Fergu-
son before the National Economists Club and Society of Government Economists
Https://www.federalreserve.gov/boarddocs/speeches/2001/20010419/default.htm.

Fernholz, Ricardo & Christoﬀer Koch. 2016. Why are big banks getting bigger? Federal Reserve

Bank of Dallas Working Paper .

Foley-Fisher, Nathan, Rodney Ramcharan & Edison Yu. 2016. The impact of unconventional
monetary policy on ﬁrm ﬁnancing constraints: Evidence from the maturity extension program.
Journal of Financial Economics 122(2). 409–429.

Frankel, Jeﬀrey A. 1982. A test of perfect substitutability in the foreign exchange market. Southern

Economic Journal 406–416.

Fratzscher, Marcel, Marco Lo Duca & Roland Straub. 2018. On the international spillovers of us

quantitative easing. The Economic Journal 128(608). 330–377.

Gertler, Mark & Simon Gilchrist. 1994. Monetary policy, business cycles, and the behavior of

small manufacturing ﬁrms. The Quarterly Journal of Economics 109(2). 309–340.

Ghironi, Fabio & Galip Kemal Ozhan. 2019. Interest rate uncertainty as a policy tool .

192

Gilchrist, Simon, Vivian Yue & Egon Zakrajšek. 2019. Us monetary policy and international bond

markets. Tech. rep. National Bureau of Economic Research.

Gilchrist, Simon & Egon Zakrajšek. 2013. The impact of the federal reserve’s large-scale asset
purchase programs on corporate credit risk. Journal of Money, Credit and Banking 45(s2).
29–57.

Greenwood, Robin, Samuel Hanson & Jeremy C Stein. 2010. A gap-ﬁlling theory of corporate

debt maturity choice. The Journal of Finance 65(3). 993–1028.

Grosse-Rueschkamp, Benjamin, Sascha Steﬀen & Daniel Streitz. 2019. A capital structure channel

of monetary policy. Journal of Financial Economics .

Guo, Xing. 2019. Financial shocks and investment ﬂuctuation: Small ﬁrms vs. large ﬁrms. Mimeo,

University of Michigan.

Gupta, Rangan, Chi Lau, Jacobus Nel, Xin Sheng et al. 2020. Monetary policy uncertainty spillovers

in time-and frequency-domains. Tech. rep.

Gürkaynak, Refet, Brian Sack & Jonathan Wright. 2007. The U.S. Treasury yield curve: 1961 to

the present. Journal of Monetary Economics 54. 2291–2304.

Gürkaynak, Refet, Brian Sack & Jonathan Wright. 2010. The TIPS yield curve and inﬂation

compensation. American Economic Journal: Macroeconomics 2. 70–92.

Gürkaynak, RS, B Sack & ET Swanson. 2005. Do actions speak louder than words? the response of
asset prices to monetary policy actions and statements. International Journal of Central Banking
1(1). 55–93.

Gürkaynak, Refet S., Brian Sack & Eric Swanson. 2005. The sensitivity of long-term interest
rates to economic news: Evidence and implications for macroeconomic models. American
Economic Review 95(1). 425–436. doi:10.1257/0002828053828446. http://www.aeaweb.
org/articles?id=10.1257/0002828053828446.

Hamilton, James. 2018. Why you should never use the Hodrick-Prescott ﬁlter. Review of Economics

and Statistics 100(5). 831–843.

Hanson, Samuel G, David O Lucca & Jonathan H Wright. 2017. The excess sensitivity of long-term

rates: A tale of two frequencies .

Hanson, Samuel G. & Jeremy C. Stein. 2015a. Monetary policy and long-term real rates. Journal

of Financial Economics 115. 429–448.

Hanson, Samuel G & Jeremy C Stein. 2015b. Monetary policy and long-term real rates. Journal

of Financial Economics 115(3). 429–448.

Hausman, Joshua & Jon Wongswan. 2011. Global asset prices and fomc announcements. Journal

of International Money and Finance .

193

Heiney, Joseph. 2010. Consolidation and proﬁtability in the US banking industry. Journal of

Business & Economics Research (JBER) 8(1).

Hennessy, Christopher & Toni Whited. 2005. Debt dynamics. The Journal of Finance 60(3).

1129–1165.

Husted, Lucas, John H Rogers & Bo Sun. 2017. Monetary policy uncertainty. FRB International

Finance Discussion Paper (1215).

Iacoviello, Matteo & Gaston Navarro. 2019. Foreign eﬀects of higher us interest rates. Journal of

International Money and Finance .

Ilzetzki, Ethan, Carmen M Reinhart & Kenneth S Rogoﬀ. 2019. Exchange arrangements entering
the twenty-ﬁrst century: Which anchor will hold? The Quarterly Journal of Economics 134(2).
599–646.

Ippolito, Filippo, Ali K Ozdagli & Ander Perez-Orive. 2018. The transmission of monetary policy

through bank lending: The ﬂoating rate channel. Journal of Monetary Economics 95. 49–71.

Istreﬁ, Klodiana & Sarah Mouabbi. 2018. Subjective interest rate uncertainty and the macroecon-

omy: A cross-country analysis. Journal of International Money and Finance 88. 296–313.

Jeenas, Priit. 2018. Monetary policy shocks, ﬁnancial structure, and ﬁrm activity: A panel approach.

Working Paper .

Jeenas, Priit. 2019. Firm balance sheet liquidity, monetary policy shocks, and investment dynamics.

Working Paper .

Jermann, Urban & Vincenzo Quadrini. 2012. Macroeconomic eﬀects of ﬁnancial shocks. American

Economic Review 102(1). 238–71.

Jordà, Òscar. 2005. Estimation and inference of impulse responses by local projections. American

Economic Review 95(1). 161–182.

Jordà, Òscar, Moritz Schularick, Alan M Taylor & Felix Ward. 2019. Global ﬁnancial cycles and

risk premiums. IMF Economic Review 67(1). 109–150.

Joslin, Scott, Kenneth J Singleton & Haoxiang Zhu. 2011. A new perspective on gaussian dynamic

term structure models. The Review of Financial Studies 24(3). 926–970.

Kabaca, Serdar. 2016. Quantitative easing in a small open economy: An international portfolio

balancing approach. Tech. rep. Bank of Canada Staﬀ Working Paper.

Kalemli-Ozcan, Sebnem. 2019. Us monetary policy and international risk spillovers. Proceed-

ings—Economic Policy Symposium—Jackson Hole, Federal Reserve Bank of Kansas City .

Kalemli-Ozcan, Sebnem, Luc Laeven & David Moreno. 2018. Debt overhang, rollover risk,
and corporate investment: Evidence from the european crisis. Tech. rep. National Bureau of
Economic Research.

194

Karabarbounis, Marios, Patrick Macnamara & Roisin McCord. 2014. A business cycle analysis of

debt and equity ﬁnancing. Economic Quarterly (1Q). 51–85.

Karceski, Jason, Steven Ongena & David Smith. 2005. The impact of bank consolidation on

commercial borrower welfare. The Journal of Finance 60(4). 2043–2082.

Khan, Aubhik, Tatsuro Senga & Julia K Thomas. 2016. Default risk and aggregate ﬂuctuations in

an economy with production heterogeneity. Unpublished Manuscript .

Kim, Don H. & Jonathan H. Wright. 2005. An arbitrage-free three-factor term structure model
and the recent behavior of long-term yields and distant-horizon forward rates. Finance and
Economics Discussion Series 2005-33 Board of Governors of the Federal Reserve System.

Kolasa, Marcin & Grzegorz Wesołowski. 2020.

Journal of International Economics 103330.

International spillovers of quantitative easing.

Kroencke, Tim Alexander, Maik Schmeling & Andreas Schrimpf. 2018. The fomc risk shift.

Available at SSRN 3072912 .

Kudlyak, Marianna & Juan Sanchez. 2016. Revisiting gertler-gilchrist evidence on the behavior of

small and large ﬁrms. FRB St. Louis Working Paper .

Kuttner, Kenneth N. 2001. Monetary policy surprises and interest rates: Evidence from the fed

funds futures market. Journal of monetary economics 47(3). 523–544.

Lakdawala, Aeimit. 2018. The growing impact of us monetary policy on emerging ﬁnancial

markets: Evidence from india. Available at SSRN 3274445 .

Lakdawala, Aeimit. forthcoming. Decomposing the eﬀects of monetary policy using an external

instruments svar. Journal of Applied Econometrics .

Lakdawala, Aeimit & Matthew Schaﬀer. 2019a. Federal reserve private information and the stock

market. Journal of Banking and Finance 106. 34–49.

Lakdawala, Aeimit & Matthew Schaﬀer. 2019b. Federal reserve private information and the stock

market. Journal of Banking & Finance 106. 34–49.

Lee, Jinsook. 2017. How do ﬁrms choose their debt types? Mimeo, Canisius College.
Li, Xinlei. 2017. Relationship lending in syndicated loans: A participant’s perspective: Columbia

University dissertation.

Lins, Karl, Henri Servaes & Peter Tufano. 2010. What drives corporate liquidity? An international

survey of cash holdings and lines of credit. Journal of Financial Economics 98(1). 160–176.

Mehrotra, Aaron N, Richhild Moessner & Chang Shu. 2019. Interest rate spillovers from the united

states: expectations, term premia and macro-ﬁnancial vulnerabilities .

Milesi-Ferretti, Gian-Maria & Cédric Tille. 2011. The great retrenchment: international capital

ﬂows during the global ﬁnancial crisis. Economic policy 26(66). 289–346.

195

Miranda-Agrippino, Silvia & Helene Rey. 2015. Us monetary policy and the global ﬁnancial cycle.

Tech. rep. National Bureau of Economic Research.

Mugasha, Agasha. 1998. Law of multi-bank ﬁnancing. McGill-Queen’s Press-MQUP.

Nakamura, Emi & Jón Steinsson. 2018. High-frequency identiﬁcation of monetary non-neutrality:

the information eﬀect. The Quarterly Journal of Economics 133(3). 1283–1330.

Neely, Christopher J. 2015. Unconventional monetary policy had large international eﬀects. Journal

of Banking & Finance 52. 101–111.

Ongena, Steven & David Smith. 2001. The duration of bank relationships. Journal of Financial

Economics 61(3). 449–475.

Ottonello, Pablo & Thomas Winberry. 2018. Financial heterogeneity and the investment channel

of monetary policy. Tech. rep. National Bureau of Economic Research.

Ozdagli, Ali. 2018. Financial frictions and the stock price reaction to monetary policy. The Review

of Financial Studies 31(10).

Petrosky-Nadeau, Nicolas & Etienne Wasmer. 2013. The cyclical volatility of labor markets under

frictional ﬁnancial markets. American Economic Journal: Macroeconomics 5(1). 193–221.

Rajan, Raghuram. 1992. Insiders and outsiders: The choice between informed and arm’s-length

debt. The Journal of Finance 47(4). 1367–1400.

Rey, Helene. 2013. Dilemma not trilemma: The global ﬁnancial cycle and monetary policy
independence. Proceedings—Economic Policy Symposium—Jackson Hole, Federal Reserve
Bank of Kansas City .

de los Rios, Antonio Diez & Maral Shamloo. 2017. Quantitative easing and long-term yields in

small open economies. International Monetary Fund.

Sharpe, Steven. 1990. Asymmetric information, bank lending, and implicit contracts: A stylized

model of customer relationships. The Journal of Finance 45(4). 1069–1087.

Swanson, Eric T. & John C. Williams. 2014. Measuring the eﬀect of the zero lower bound on
medium- and longer-term interest rates. American Economic Review 104(10). 3154–85. doi:10.
1257/aer.104.10.3154. http://www.aeaweb.org/articles.php?doi=10.1257/aer.104.
10.3154.

Whited, Toni M & Guojun Wu. 2006. Financial constraints risk. The Review of Financial Studies

19(2). 531–559.

Wu, Wenbin. 2018. The credit channel at the zero lower bound through the lens of equity prices.

Journal of Money, Credit and Banking 50(2-3). 435–448.

Zetlin-Jones, Ariel & Ali Shourideh. 2017. External ﬁnancing and the role of ﬁnancial frictions

over the business cycle: Measurement and theory. Journal of Monetary Economics 92. 1–15.

196