Xu Dissertation 2019

Essays on Finance in History
Citation
Xu, Chenzi. 2019. Essays on Finance in History. Doctoral dissertation, Harvard University,
Graduate School of Arts & Sciences.
Permanent link
http://nrs.harvard.edu/urn-3:HUL.InstRepos:42029827
Terms of Use
This article was downloaded from Harvard University’s DASH repository, and is made available
under the terms and conditions applicable to Other Posted Material, as set forth at http://
nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-use#LAA
Share Your Story

The Harvard community has made this article openly available.
Please share how this access benefits you. Submit a story .
Accessibility
A dissertation presented
by
Chenzi Xu
to
The Department of Economics
in partial fulfillment of the requirements

for the degree of
Doctor of Philosophy
in the subject of
Economics
Harvard University
Cambridge, Massachusetts
April 2019
c 2019 Chenzi Xu
All rights reserved.

Dissertation Advisors: Author:
Professor Jeremy Stein Chenzi Xu
Professor Kenneth Rogoff
Abstract
This dissertation presents three essays on financial markets in various historical contexts
to assess their role in international trade, private money creation, and safe assets creation,
respectively. Chapter 1 shows that a temporary financial sector shock can have long-lasting
consequences for the patterns of international trade. Using the most severe banking crisis
in British history as a laboratory, I find that countries more exposed to the bank failures
in London had significant losses in exports market share for four decades, and that the
effects were most severe in countries with higher contractual frictions and limited access
to alternative forms of financing. Chapter 2, joint with he Yang, studies a policy change
in the United States, the National Banking Act of 1864, that improved the quality of the
liabilities issued by federally regulated national banks. Using an exogenous change in the
cost of gaining access to a national bank, we find that the improvement in the liabilities’
usefulness as a money-like asset had strong effects on a local economy’s commercial
and trade activities. Chapter 3 presents evidence demonstrating that a demand for safe
assets issued by the central country is a feature of the international monetary system
that predates the modern era. Using institutional and empirical evidence from the pre-
WWI classical gold standard era, I show that investors were willing to pay a premium
for holding the economy’s safest asset. In addition, the historical setting allows me to
instrument for the supply of safety using shocks to gold mining stemming from weather
fluctuations in South Africa.
iii
Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
0 Preface 1
1 Reshaping Global Trade: The Immediate and Long-Run Effects of Bank

Failures 3
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 London’s banks: institutional & historical context . . . . . . . . . . . . . . . 12
1.2.1 Trade finance & British banking dominance . . . . . . . . . . . . . . 12
1.2.2 London banking crisis of 1866 . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Measuring exposure to London’s crisis around the world . . . . . . . . . . . 16
1.3.1 Measuring the shock to bank credit . . . . . . . . . . . . . . . . . . . 17
1.3.2 Validity of reduced form estimation . . . . . . . . . . . . . . . . . . . 18
1.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4.1 Exposure to bank failures . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4.2 Immediate outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.4.3 Long-term outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.5 Immediate impact on trade . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.5.1 Intensive margin effect: baseline specification . . . . . . . . . . . . . 32
1.5.2 Extensive margin effects . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.5.3 Limited within-country substitution . . . . . . . . . . . . . . . . . . . 43
1.6 Long-term impact on trade . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
1.6.1 Baseline results across countries . . . . . . . . . . . . . . . . . . . . . 47
1.6.2 Channels for persistence . . . . . . . . . . . . . . . . . . . . . . . . . . 55
1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2 National Banks and the Liabilities Channel of Local Economic Activity 63

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.2 Historical context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
iv
2.2.1 The Free Banking Era . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
2.2.2 The National Banking Era . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.3 Data & empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.3.1 Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.3.2 Instrument for national bank entry . . . . . . . . . . . . . . . . . . . 81
2.3.3 Pre-period balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.3.4 First stage results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.4.1 Changes in the agriculture sector . . . . . . . . . . . . . . . . . . . . . 92
2.4.2 Trade activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.4.3 Growth in the manufacturing sector . . . . . . . . . . . . . . . . . . . 101
2.4.4 Growth in innovation . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
2.4.5 Long-term effects and spillover effects . . . . . . . . . . . . . . . . . . 110
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
3 Safety in the Gold Standard Era 116

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.2 Institutional context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3.2.1 The functioning of the gold standard . . . . . . . . . . . . . . . . . . 119
3.2.2 Private money creation and the supply of safety . . . . . . . . . . . . 121
3.2.3 South Africa mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.3.1 Finance data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.3.2 South African data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.4 Demand for safe assets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.4.1 Main results using OLS . . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
References 132
A Appendix to Chapter 1 142

A.1 Additional tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
A.2 Additional figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
A.3 Instrument validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
A.4 Additional evidence on long-term effects . . . . . . . . . . . . . . . . . . . . 174
A.4.1 Lower-cost financing for shorter routes . . . . . . . . . . . . . . . . . 174
A.5 Additional historical context . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
A.5.1 Trade finance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
v
A.5.2 Overend & Gurney . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
A.5.3 London banking crisis . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
A.5.4 Country characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . 185
A.6 Historical data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
A.6.1 Data constructed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
A.6.2 Data collected . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
B Appendix to Chapter 2 191

B.1 Additional tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
B.2 Additional figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
B.3 Additional notes on the first stage . . . . . . . . . . . . . . . . . . . . . . . . 199
B.3.1 First stage regressions in alternative samples . . . . . . . . . . . . . . 199
B.3.2 Timing of entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
B.4 Additional historical context . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
B.4.1 Banking system before the National Banking Act . . . . . . . . . . . 201
B.4.2 Additional evidence of bank debt illiquidity in the Free Banking Era 202
vi
List of Tables
1.1 Pre-crisis comparison of bank characteristics . . . . . . . . . . . . . . . . . . 21

1.2 Correlation between bank failures and pre-crisis location characteristics . . 24
1.3 Summary statistics: Ports and countries . . . . . . . . . . . . . . . . . . . . . 31
1.4 Immediate effect of bank failures on port-level shipping . . . . . . . . . . . 36
1.5 Immediate effect of bank failures on country-level values of exports . . . . 42
1.6 Extensive margin effect of exposure to bank failures . . . . . . . . . . . . . . 44
1.7 Immediate effect of bank failures on port- and country-level shipping . . . 46
1.8 Port access to alternative sources of financing . . . . . . . . . . . . . . . . . 56
2.1 Comparison of National Banks and state banks . . . . . . . . . . . . . . . . 76

2.2 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.3 Pre-period covariate balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.4 First stage regressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
2.5 National Banks and local agricultural sector (1880-1890) . . . . . . . . . . . 94
2.6 National Banks and the shift in agricultural production . . . . . . . . . . . . 95
2.7 National Banks and local business (town-level) . . . . . . . . . . . . . . . . . 97
2.8 National Banks and local business (county-level) . . . . . . . . . . . . . . . . 99
2.9 Price levels and existence of National Banks . . . . . . . . . . . . . . . . . . 101
2.10 National Banks and growth in manufacturing production . . . . . . . . . . 102
2.11 National Banks and growth in manufacturing inputs, employment, and
capital . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.1 Relationship between supply of safety (gold) and safety spread at different
maturities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
A1 Bank-level relationship between failure and credit supply . . . . . . . . . . 142

A2 Bank balance on geographic exposure calculated as percent of assets . . . . 143
A3 Robustness to removing cotton exporting countries: immediate effect of
exposure to bank failures on port-level shipping . . . . . . . . . . . . . . . . 144
vii
A4 Robustness to controls: immediate effect of exposure to bank failures on
country-level shipping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
A5 Robustness to different clustering: immediate effect of exposure to bank
failures on country-level exports . . . . . . . . . . . . . . . . . . . . . . . . . 146
A6 Elasticity of trade to physical distance . . . . . . . . . . . . . . . . . . . . . . 147
A7 Long-term effects of financing shock on country-level exports . . . . . . . . 148
A8 Effect of bank failures from 1850–2014 . . . . . . . . . . . . . . . . . . . . . . 149
A9 Long-term effects: robustness to gravity measures of commonality . . . . . 150
A10 Long-term effects: robustness to monetary standard and conflict . . . . . . 151
A11 Long-term effects: robustness to industry composition of exports . . . . . . 152
A12 Long-term effects: robustness to excluding cotton exporting countries . . . 153
A13 Long-term effects: robustness to contemporaneous financial crises . . . . . 154
A14 Long-term effects: robustness to financial crises in 1865 . . . . . . . . . . . . 155
A15 Long-term effects: robustness to borrowing from London Stock Exchange . 156
A16 Long-term effects: robustness to composition of banks . . . . . . . . . . . . 157
A17 Examples of banks and operating regions . . . . . . . . . . . . . . . . . . . . 178
A18 Countries by region and British empire status . . . . . . . . . . . . . . . . . 185
B1 Alternative sample periods for first stage regressions . . . . . . . . . . . . . 191

B2 Alternative samples for first stage regressions . . . . . . . . . . . . . . . . . 192
B3 Placebo: National Banks and local business (county-level) . . . . . . . . . . 193
B4 Robustness: National Banks and growth in manufacturing production . . . 194
B5 Placebo: National Banks and local manufacturing sector (1870-1880) . . . . 195
B6 Placebo: National Banks and local innovation activity (1870-1880) . . . . . . 195
viii
List of Figures
1.1 Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1.2 Geography of banking and trade . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.3 Correlation between exposure to bank failures and shipping in 2 year window 34
1.4 Aggregate exports, grouping countries by above and below average expo-
sure to bank failures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
1.5 Financing shock has long-term effects on exports . . . . . . . . . . . . . . . 51
1.6 Recovery is better with access to other banks . . . . . . . . . . . . . . . . . . 58
1.7 Recovery is worse within groups of countries with similar exports . . . . . 60
2.1 Relative discounts on state bank notes relative to Philadelphia . . . . . . . . 73

2.2 National and state banks: (1863-1900) . . . . . . . . . . . . . . . . . . . . . . 77
2.3 Exhibits of data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.4 Histograms of town population in 1880 . . . . . . . . . . . . . . . . . . . . . 84
2.5 Maps of towns in the sample . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.6 Conditional covariates balance . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2.7 Production per capita and innovation . . . . . . . . . . . . . . . . . . . . . . 109
2.8 Persistent positive effect on manufacturing production . . . . . . . . . . . . 112
2.9 Spillover effect on neighboring counties . . . . . . . . . . . . . . . . . . . . . 113
3.1 Banking Department balance sheet . . . . . . . . . . . . . . . . . . . . . . . . 120

3.2 Composition of Assets in the Banking Department (1870–1914) . . . . . . . 124
3.3 Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.4 Relationship between supply of Gold/GDP and safety premium . . . . . . 129
A1 Industry composition of global exports in 1865 . . . . . . . . . . . . . . . . . 158

A2 Distribution of exposure to bank failure . . . . . . . . . . . . . . . . . . . . . 159
A3 Positive correlation between country-level number of ships and exports
values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
A4 Port-level effect of bank failures on exports: robustness to distance cutoffs . 160
ix
A5 Positive correlation between sailing distance and geodesic distance . . . . . 160
A6 Positive correlation between news lag and geodesic distance to London . . 161
A7 Effect of above average exposure to bank failure on total exports . . . . . . 162
A8 Annual country-level growth rates of total exports . . . . . . . . . . . . . . . 163
A9 Aggregate GDP, grouped by above and below average exposure to bank
failures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
A10 Effect of bank failures from 1850–2014 . . . . . . . . . . . . . . . . . . . . . . 164
A11 Growth of multinational banks, 1850-1913 . . . . . . . . . . . . . . . . . . . . 165
A12 Banking sector recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
A13 Effect of exposure to bank failure on new vs pre-existing trade relationships167
A14 Treatment placebo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
A15 Exports correlation within country regions . . . . . . . . . . . . . . . . . . . 168
A16 Country region placebo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
A17 Effect of other-country exposure within region on own country . . . . . . . 170
A18 Recovery within country groups . . . . . . . . . . . . . . . . . . . . . . . . . 171
A19 Exports are not affected for closer destinations . . . . . . . . . . . . . . . . . 175
A20 Bank of England Discount Window lending in 1866 . . . . . . . . . . . . . . 182
B1 Bank note from the Pittsfield Bank in Massachusetts . . . . . . . . . . . . . 196

B2 Distribution of national banks in 1885 . . . . . . . . . . . . . . . . . . . . . . 197
B3 Timing of entry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
x
Acknowledgments
I am very grateful for the support and insight from my advisors Jeremy Stein, Kenneth
Rogoff, Claudia Goldin, Matteo Maggiori, Sam Hanson, Adi Sunderam, and Nathan
Nunn. To Jeremy I owe the encouragement to continue pursuing the project that would
eventually become my job market paper. To Jeremy, Ken, and Claudia, I am thankful for
their unwavering presence and tremendous examples of scholarly dedication. Without
Ken and Claudia, whose guidance I am lucky to have had since my undergraduate years,
I would not have done this PhD. Matteo, Sam, Adi, and Nathan, provided invaluable
advice, and I have learned so much from them.
I could not have asked more from my friends both in and out of Littauer over these
years. Thanks to Josh Abel, Laura Blattner, Val Bolotonyy, Kirill Borusyak, Alonso de
Gortari, Ellora Derononcourt, Andy Garin, Siddarth George, Pika Goldin, Ben Golub,
Elizabeth Guo, Leander Heldring, Xavier Jaravel, Ian Kumekawa, Lawrence Li, Libby
Mishkin, Heather Sarsons, Nihar Shah, Jann Spiess, and He Yang.
I owe the most to my family–my parents Mingguang Xu and Xiuqin Xia, my sister
AliAnn Xu, and my partner Chris Bridge. I love them dearly.
xi
To my family.
xii
0 | Preface
This dissertation comprises of three independent chapters in the areas of economic history
and finance that are tied together by the themes of persistence, money, and credit.
All three chapters are historical, and economic history provides the lens for examining
the first theme of persistence. Persistence is a topic that is uniquely suited to being
studied by economic historians. In this dissertation, persistence refers to two types of
phenomenon: the first is the potentially long-lasting effects from events or shocks, and the
second is institutions or events that recur despite differences in time periods or settings.
The first type of persistence can only be seen through the long lens of history, in which
time tells us why and how long certain changes lasted. The second type of persistence
lends us insight on “fundamental” characteristics of economies.
In chapter one, I capture the persistent effects of a temporary shock to financial
markets on patterns of international trade. This type of persistence can be viewed as
either a slow transition path from one equilibrium to the next, or else a shift to a new
equilibrium completely. In the second view of persistence as recurrence, I focus on
events or institutional characteristics that re-emerge over time despite disruptions (or
abatements). Financial crises (the topic of my first chapter) and the institutional traits of
the international monetary system (the topic of my third chapter) are both persistent in
this sense.
Finance provides the lens for the themes of money and credit. Money, money-ness, and
the real economic consequences of having access to money are themes that I explore in
1
chapters two and three. In particular, I focus on private money creation from the liabilities
issued by financial institutions. While much of the literature on these instruments has
focused on their negative externalities from a risk-shifting perspective or in crisis states, I
show how two historically significant institutions (such as the Office of the Comptroller
of the Currency in the United States and the Bank of England in the United Kingdom)
implemented policies that guaranteed the safety of these liabilities, and hence positively
affected commercial transactions.
The final theme is credit. Financial crises are often told as stories about shocks to
credit: bank failures and debt defaults lead to loss of access to both short-term and
long-term credit markets. The crisis at the center of my study in chapter 1 led to a severe
contraction in the short-term trade credit that exporters relied on for international trade.
Although credit recovered within a few years, the patterns of trade were disrupted for
four decades. In this case, while lack of access to credit is the precipitating force in the
persistent effects, recovery in credit is not itself enough to restore the original equilibrium.
In chapter 2, while it is not possible to fully rule out the changes to credit provision after
the National Banking Act, we argue that the nature of lending (on the asset side of the
balance sheet) did not change nearly as much as the nature of transactions (operating
through the liabilities side of the balance sheet). In both these chapters, while credit is
relevant, it does not tell the full story, and the study of finance benefits from examining
both what it can and cannot explain.
2
1 | Reshaping Global Trade:
The Immediate and Long-Run
Effects of Bank Failures
1.1 Introduction
Banking crises have occurred repeatedly in countries across the income spectrum through-
out history, and a recent empirical literature has shown that they have severe consequences
for short-term real economic activity.1 Models of the macroeconomic response to financial
sector disruptions typically imply that recovery in the health of the banking sector will
lead to recovery in the real economy (Bernanke and Gertler, 1989; Kiyotaki and Moore,
1997). However, the short-term adjustments triggered by banking crises appear to have
longer lasting economic consequences (Cerra and Saxena, 2008). International trade is a
sector that is both sensitive to the costs of external finance (Amiti and Weinstein (2011);
Paravisini et al. (2014)) and could theoretically exhibit path dependence, where a one-time
temporary shock leads to a persistent change in the composition of exporters (Baldwin,
1988; Baldwin and Krugman, 1989). Yet establishing the causal effect of bank failures on
exports beyond the level of the firm is difficult because local conditions simultaneously
1 Recent work includes Chodorow-Reich (2014), Benmelech et al. (2016), Huber (2018) on employment,
and Ashcraft (2005), Richardson and Troost (2009), Calomiris and Mason (2003), Frydman et al. (2015) on
investment and output.
3
impact economic activity and banking sector health. Even when it is possible to isolate an
exogenous shock to the domestic banking sector, studies have been limited to examining
short-term outcomes within one country.
This paper causally estimates the impact of bank failures on international trade in
a unique historical setting that lends global coverage and makes it possible to study
long-term effects. The laboratory is the natural experiment arising from the most severe
financial crisis in British history, the 1866 London banking crisis, which occurred when
London was the center of the global financial system and British multinational banks
were the dominant providers of trade credit around the world. I show that the crisis
disrupted the normal flow of credit. Using the bank-level shocks from the crisis, I find
that this temporary shock had both immediate and long-lasting effects on international
trade patterns.
The 1866 crisis was caused by the unexpected bankruptcy of the fraudulent financial
market intermediary Overend and Gurney. Its announcement of bankruptcy led to panic
and severe bank runs on all London banks. Crucially, Overend and Gurney was not
itself involved in trade finance or trade-related activities.2 However, in the immediate
aftermath, 12 percent of British multinational banks (weighted by size) failed.3 These
multinational banks borrowed funds in London and lent them abroad through subsidiary
offices in cities around the world. Headquarter failures in London severed this funding
structure and necessitated that all foreign operations stop as well. Port cities and countries
around the world differed in their pre-crisis dependence on the British banks that failed
and were therefore differentially exposed to the crisis in London.
Several features of the historical setting make it well-suited for identifying the causal
effect of exposure to bank failures on exporting activity. First, British multinational
banks were dominant and had global reach: they provided over 90 percent of trade
credit in cities around the world, and they operated in countries that accounted for 98
2 See appendix A.5.2.
3 There were 128 multinational banks, of which 22 failed.
4
percent of world exports in 1865.4 Their competitive advantage relative to other local,
or even European alternatives, stemmed from their unique structure of lending abroad
but drawing funding from the largest money market in the world. Subsidiary locations
dependent on British banks paid a lower cost of capital on average but were exposed to
fluctuations in the cost of credit from London. This structure of global operations meant
that a single shock in the international financial center—the failure of London’s largest
financial market intermediary—impacted banking activity around the world.5
Second, these multinational banks were chartered to provide trade credit, which
establishes a natural link between their operations and exporting activity (Baster, 1934).
The banks’ similarities in funding and management structure also makes it likely that they
affected exports through the same channels across locations. Third, outside of Britain,
there was no post-crisis government or policy intervention in the macroeconomy, so the
estimated effects are not conditional on the degree of the response.6 Finally, the 1866 crisis
was followed by almost five decades of relative global peace, one of the longest in modern
history, when both goods and capital flows faced few barriers, in what was known as
the First Age of Globalization (O’Rourke and Williamson, 1999). Together, these features
allow me to empirically isolate the effect of bank failures from other determinants of local
economic development and to examine the process of recovery over many decades.
In order to conduct the empirical analysis, I construct several new datasets of historical
trade and financing activity around the world, at the port city and country levels. First,
I collect over 11,000 handwritten loan contracts from archival records comprising the
universe of pre-crisis British bank lending relationships in cities around the world. To my
knowledge, these are the only data with full global coverage of the dominant financial
4 Author’s calculations based on the locations and operations of British banks and non-British banks,
and the value of exports across countries in 1865.
5 International capital reversals continue to cause cross-border contagion today. Cetorelli and Goldberg
(2011); Huber (2018); Iyer et al. (2013); Paravisini et al. (2014); Peek and Rosengren (1997, 2000); Puri et al.
(2011); Schnabl (2012) study their effects in a variety of different national contexts.
6 Romer and Romer (2018) document that in the post-Bretton Woods era, the output decline following
financial crises is highly dependent on policymakers’ ability to enact post-crisis countercyclical policies.
5
center’s banking relationships in any time period, and they make it possible to causally
link a single shock to outcomes around the world. Second, I quantify city-level exporting
activity in the short-term using a dataset of comprehensive shipping activity in port cities
around the world built from the daily Lloyd’s List newspaper.7 Third, I build a panel
of exporting and financing activity at the country- and city-levels, respectively, from
1850–1914.
To identify the causal relationship between bank failures and exporting activity, I use a
difference-in-difference (DD) estimator with continuous treatment intensity, allowing for
a control group of places with no exposure to British banks in 1866. I measure a location’s
exposure to bank failures as the fraction of its credit pre-crisis that came from the banks
operating in that location that failed, where locations are cities and countries, respectively.
This measure follows a Bartik/shift-share structure of exposure to bank-health shocks
used in Greenstone et al. (2014), Chodorow-Reich (2014), and Amiti and Weinstein (2018)
among others at the firm-level.8 This strategy is based on the theoretical and empirical
evidence that banks lend locally since contractual frictions between banks and their
borrowers increase with distance (Mian, 2006; Petersen and Rajan, 2002; Sharpe, 1990).
Identification relies on there not being a simultaneous shock to a location that would
cause both its exports to decline and the banks operating there to fail. First, I show that
bank failures are uncorrelated with observable characteristics of the banks themselves,
which helps to address the key endogeneity concern that riskier banks sorted to locations
that would have experienced exports declines anyways. Consistent with the environment
of limited knowledge during the 1866 panic, worse banks (proxied by observable pre-
crisis balance sheet characteristics) did not experience more severe runs and were not
more likely to fail. Second, and more importantly, I verify that bank failures are mostly
7 Annual country-level shipping activity is highly correlated with annual values of exports, and I verify
my short-term findings at the country-level using values of exports.
8 Several contributions to this literature also estimate the within-firm effects using connections to multiple
borrowers (Amiti and Weinstein, 2018; Blattner et al., 2018). However, the within-firm variation is only
useful when outcomes are also at the bank-firm level (see Khwaja and Mian (2008)). In my analogous
exercise using location-level effects, exports are not observed at the bank-location level.
6
uncorrelated with observable characteristics of subsidiary locations, for example of their
value of exports, specialization in particular commodities, or military conflicts. The
lack of location-level correlations with bank failures helps to address the endogeneity
concern that these characteristics were the proximate cause of both bank failures and
exports declines. Finally, I control for potentially confounding observable factors and
include a number of robustness checks to provide additional evidence for the identifying
assumption.
My analysis proceeds in the following way. First, I examine whether this finance-driven
shock to trade costs lowered exporting activity in the short-term. Second, I establish
that the disruption to finance was temporary: cities more exposed to bank failures in
1866 had access to the same number of banks as less-exposed cities by 1871.9 Third, I
examine the long-run effects: if exporting activity primarily depended on short-term
financing, recovery should follow. However, if the temporary financing shock severely
disadvantaged exporters during this period of massive growth in global trade, then the
initial loss of market share abroad would lead to persistently lower levels of exports in
the long-run.
I find that exposure to the failure of these multinational banks caused large and
immediate contractions on both the intensive (the amount exported) and extensive
(whether they exported at all) margins of exporting activity within and across countries.
Ports exposed to a 10 pp increase in bank failures shipped 5.6 percent less the year
following the crisis compared with unexposed ports within the same country. The
intensive margin findings are larger than Amiti and Weinstein (2011) and Paravisini et al.
(2014)’s firm-level results in more recent settings. The difference is most likely because
bank failure is more extreme than declines in bank health and less prone to measurement
error. In addition, I find extensive margin losses in the number of exports destinations
9I measure access to banks by building a dataset of the universe of city-level multinational bank
subsidiary operations around the world in five-year windows from 1850-1914. I count banks of all
nationalities.
7
and the likelihood that a port traded at all. These results are consistent with findings in
modern data documenting that credit constraints have a negative impact on firm entry
into exporting (Berman and Héricourt, 2010).
At the country-level, I estimate an even larger loss in shipping activity from exposure to
bank failures, suggesting that general equilibrium forces did not substantially reallocate
exporting activity within the exporting country. I explicitly test for any short-term
reallocation by estimating the response of port-level shipping to the average level of bank
failure exposure in other ports within a country, controlling for each port’s own exposure.
The effect is not significantly different from zero. Since ships are highly mobile across
ports, this result suggests that the binding friction was the costly process of forming new
lending relationships (Bernanke, 1983; Rajan, 1992).
Having established the short-term effect of this financing shock on exporting activity,
I show that the losses across countries persisted for decades in the aggregate and in
terms of market share despite fast recovery in the banking sector. After 1866, there is
an immediate and permanent divergence in the aggregate levels of exports between
countries with above average exposure to bank failures compared to those with below
average exposure. This divergence is driven by an increase in the growth rate of exports
for less exposed countries right after the crisis.10 Benchmarked against estimates of the
elasticity of trade to physical distance, exposure to an above average bank failure shock
after two years is equivalent to a 30.5 percent increase in a country’s distance to its trade
partners. These initial differences lead to a 1.8 percent difference in the average annual
growth rate of exports from 1866 until 1914.11
In order to estimate the market share effect, I extend my short-term DD identification
strategy and estimate the cross-sectional elasticity of country-level values of exports
10 The main analysis stops in 1914 because of the economic and institutional upheavals of WWI. However,
the divergence persists until 2014, indicating that countries more exposed to bank failures never experience
a compensating positive growth shock.
11 The partial equilibrium framework I have adopted can only estimate counterfactual declines in global
trade volumes assuming that there is no across-country exports substitution. However, I provide evidence
of high exports substitutability, leading to a shift in the patterns of global trade if not the global levels.
8
to bank failures in every year in a dynamic DD. This estimator compares the relative
amounts imported by a given country in a given year from exporters exposed to varying
degrees of bank failure, controlling for bilateral measures of geographical and institutional
distance.12 Incorporating the bilateral resistance measures means this estimator takes the
form of a fixed effects estimation of a general structural gravity equation used to quantify
the responsiveness of exports to trade costs (Head and Mayer, 2014). I find that there
are no differential pre-trends from 1850–1866, a large negative effect beginning in 1867,
and statistically significant differences in exports market share until 1900. I also show
that the patterns of persistence and recovery cannot be explained by random divergence
among countries over time and are robust to a large number of alternative explanations.
While the estimated effects are not statistically significant after 1900, the point estimates
are persistently negative, and the magnitude in 2014 is still 53 percent of the average
magnitude from 1866–1900, indicating a very slow process of convergence.
The path dependence in exports patterns lends empirical evidence to the possibility
of multiple equilibria in the geographic distribution of economic activity (Allen and
Donaldson, 2018; Bleakley and Lin, 2012; Davis and Weinstein, 2002; Kline and Moretti,
2014a; Redding and Sturm, 2008). While the literature has focused on the role of physical
capital and geographic characteristics in determining initial conditions, to my knowledge
this paper is the first to show that temporary shocks to financial capital can be the
proximate source of divergence. The persistent losses that I document are consistent
with a framework of high substitutability across exporters (a country-level analogue
to the homogeneous firms in Baldwin and Krugman (1989)), which is plausible in this
institutional setting in which the vast majority of trade was in raw commodity goods.
Next, I explore two mechanisms for the long-term losses: an exporter’s lack of access
to alternative sources of trade financing and an importer’s ability to substitute to other
exporters selling similar products. Without ruling out the possibility of concomitant
12 Theimporter-year fixed effects control for all country-level shocks experienced by the importer, such as
aggregate demand or income shocks.
9
factors, I explore two mechanisms. First, to proxy for the availability of alternate financing
sources that could be accessed after the crisis, I use the number of non-British banks
operating in each port city, and find that access to other banks alleviates one third of the
baseline short-term reduction in exporting. In the long-term, exporters with access to
alternative banking networks were almost completely shielded from the cost of the initial
exposure to British bank failures. Second, I compare relative recovery rates within groups
of countries exporting similar goods. Countries facing more competition in exports
markets did not have any recovery in their bilateral trade relationships by 1914. As
a placebo, countries within random groupings followed the same baseline patterns of
recovery by the 1900s.
This paper shows that there are immediate and long-term global consequences from
disruptions to the dominant financial market, and it is related to a number of literatures.
In the modern economy, credit conditions in peripheral countries have been found to be
disproportionately associated with capital flows from the United States (Eichengreen and
Rose, 2004; Gourinchas et al., 2012). Rey (2015) shows that the ultimate source of these
credit cycles may be monetary policy transmitted through global banks. Separately, there
is a large literature on the Bank of England’s policies during its pre-WWI hegemony that
highlights its influence over the pound sterling (Bagehot, 1873; Flandreau and Ugolini,
2013; Schwartz, 1987). This paper empirically joins these two strands of literature to
concretely illustrate how the conditions in the dominant financial market affect real
activity globally.
Methodologically, I use quasi-random variation in bank failure at the location-level,
analogous to the firm-level measures of exposure used in recent studies (Frydman et al.,
2015; Gan, 2007). In particular, my strategy is similar to studies in which the shocks to the
domestic banking sector originated abroad (Peek and Rosengren, 2000; Puri et al., 2011;
Schnabl, 2012). While I find that real economic activity contracted even in the historical
setting, I also estimate these effects in the macroeconomy, beyond the level of the firm,
10
and in the long-term across all countries.
A separate literature has been able to correlate domestic banking crises with deep,
persistent output declines across countries (Bordo and Haubrich, 2010; Cerra and Saxena,
2008; Kaminsky and Reinhart, 1999; Krishnamurthy and Muir, 2017; Reinhart and Rogoff,
2009a; Schularick and Taylor, 2012). My estimation establishes this relationship causally.
In contrast to the multi-country studies, I focus on one crisis, which provides a single
institutional context and a clear interpretation of the role of banks within it. Using one
setting also avoids the difficulties of comparing very different shocks across countries and
time (Romer and Romer, 2017). While the Global Financial Crisis has also shown that
crises originating in the core are not just of historical interest, comparable data on the
bank linkages in 2008 are not available, and it would only be possible to observe effects
for one decade.
Finally, this paper’s focus on exports speaks to the growing literature on the role of
finance in trade. There has been revived interest in this topic following the Great Trade
Collapse of 2008, but the existing literature has not reached a consensus. Most studies
use the cross-industry variation in external finance dependence from Rajan and Zingales
(1998a) and measure a firm’s access to finance from firm balance sheets (e.g. Chor and
Manova (2012); Iacovone and Zavacka (2009)), while others adopt a structural approach
(e.g. Alessandria et al. (2010); Eaton et al. (2016)). Their findings vary from finding large
to insignificant effects.13 In contrast, I directly observe the trade financing constraint from
bank-level shocks as in Amiti and Weinstein (2011) and Paravisini et al. (2014), and I find
strong support for the financing channel. In addition, I also find a much larger decline in
trade relative to output, consistent with patterns in the modern data.
13 Ahn et al. (2011) posits the range of conclusions could stem from inconsistent or incorrect measurement
of trade credit, especially when proxied by standard measures of external finance dependence. In addition,
Feenstra et al. (2014) notes that trade finance acts through different mechanisms from standard external
financing.
11
1.2 London’s banks: institutional & historical context
This section provides an overview of the institutional structure of British multinational

banks and international trade finance in the 19th century. It describes the events leading
up to the London banking crisis of 1866 and the consequences of the firm Overend and
Gurney’s failure.
1.2.1 Trade finance & British banking dominance
Contractual frictions were a major barrier to establishing international trading relation-

ships in the 19th century, just as they still often are today (Antràs and Foley, 2015; Auboin,
2012). Due to the long lag between the initial shipment by exporters, the receipt of goods
by importers, and their final sale by importers, purchase and payment was staggered,
and there was room for default on both sides. Importers were not willing to directly
finance exporters (through cash-in-advance payment) when the exporter was risky and
losses were unlikely to be recouped. These contractual frictions were particularly high
for exporters in countries of low institutional quality or in new markets.14 Exporters
waiting for payments faced higher working capital costs, and contemporary 19th century
accounts indicate that uncertainty over payments restricted many firms from expanding
to new markets (Mackenzie, 2013; Reber, 1979).
Banks overcame these frictions by directly financing exporters during the period of
shipment. British multinational banks operated locally through subsidiary offices, which
allowed them to build long-term relationships and gave them superior knowledge of an
exporter’s risk. These offices conducted the banks’ business of lending via short-term,
often collateralized, loans called “banker’s acceptances.” Banker’s acceptances were
14 Antràs and Foley (2015) presents a symmetric case where the contractual friction could bind for either
the importer or exporter. Empirically, they study a modern US-based exporter whose partners varied in
their contractual quality. However, in the historical context, the financing friction did not bind for importers
because lending to finance purchases was not the norm.
12
a special form of a bill of exchange, which was a general debt obligation that could
be written between any two parties. Bills of exchange had the feature of joint liability,
meaning that in the case of default by the original debtor, the “acceptor” (in this case
the bank) was liable for the debt. This feature transformed the bills from bearing the
idiosyncratic risk of the individual exporter into bearing the bank’s credit risk instead.
In addition, British multinational banks had accounts at the Bank of England, which
promised to lend against collateral guaranteed by its customers at the Bank’s Discount
Window. The term “Discount Window” comes from the transaction of “discounting” bills
of exchange that took place there. Discounts most resembles a modern-day repurchase
agreement: the seller received the face value of the bill minus the discount rate (haircut) at
the initiation of the transaction, and he paid the full face value in return for the security at
its maturity. At maturity, the bill was presented to the original borrower via his accepting
bank for repayment, and the debt was terminated.15
The bills accepted by British multinational banks and implicitly guaranteed by the
Bank of England were useful debt and investment instruments, analogous to short-term
Treasury bills today.16 Banker’s acceptances were flexible and customizable, so in theory
they could be accepted for any debt obligation. However, the British multinational banks
at the center of this study used them to finance international trade. The relationship these
banks had to the Bank of England and the London money market allowed them to form
the backbone of international trade finance in the 19th century.
Three institutional details are relevant for interpreting the effect of British multinational
bank failures on trade. First, they were chartered to only fund trade and were not
permitted to act as commercial banks and invest in long-term, illiquid assets (Chapman,
1984; Muirhead and Green, 2016). Second, contemporaries emphasized that British banks
were not limited to funding trade with Britain, and in fact were integral for trade that had
15 Flandreau and Ugolini (2013) and Anson et al. (2018) study the rules governing the Bank of England’s
discounting activity during different 19th century banking crises.
16 The modern Treasury bill was proposed by Walter Bagehot in 1877 and modeled after these commercial
bills to allow the government to borrow at short maturities in a similar manner (BOE, 1964).
13
no British counterparties (Baster, 1934; Jenks, 1927).17 Third, the safe and liquid features
of their bills meant that banks could remit them back to their London headquarters which
then resold them on the London money market.18
Access to London was integral to subsidiary office operations and provided British
multinational banks with two advantages over local, and even other European, banks:
remitting bills back to London freed up local capital for more acceptances, and the
headquarters issued stock and deposits in London, where the cost of capital was low,
to lend abroad. These two factors contributed to British banking dominance and global
reach. In 1866 on the eve of the London banking crisis, the countries that British banks
operated in accounted for 98 percent of the value of global exports. A conservative
estimate is that these banks provided 91 percent of the trade credit in a given city.19 To
my knowledge, this paper is the first to study the international implications of the 1866
London banking crisis.
1.2.2 London banking crisis of 1866
The 1866 banking crisis was the largest ever shock to the London money market, when 22
out of 128 multinational banks headquartered in London (12% of banks by size) failed.20
The closures of the headquarters in London necessitated that subsidiary operations abroad
close as well, which constricted the supply of credit in subsidiary locations.
The 1866 crisis was caused by the unanticipated bankruptcy of the firm Overend and
17 Jenks (1927) writes on p. 69, “[American imports of] wines from France, coffee from Brazil, sugar from
the West Indies, and silk from Hong Kong were paid alike with bills on London.”
18 In London, the short-term funds circulated among banks who bought/sold securities to generate their
preferred maturity distribution, members of the London Stock Exchange who borrowed from banks to
purchase bonds, and interbank lenders who facilitated the transactions (Nishimura et al., 2012, p.18).
19 These two figures are the author’s own calculations. The first is based on the bank-city lending
relationships. The second assumes that non-British banks were the same size as British banks. Since British
banks were almost certainly larger than non-British banks, this figure is a lower bound to the amount of
trade credit they supplied. This circle of funding made the business of international banking potentially
very profitable, and Kisling (2017) documents that German banks began entering this market in the later
part of the 19th century to compete with British dominance.
20 This was also the last time there were bank runs in the United Kingdom until 2008. Anna Schwartz
referred to the 1866 crisis as the “Last English Financial Crisis” (Schwartz, 1987).
14
Gurney, the largest and most prestigious interbank lender in the City of London. Its
business was buying and selling liquid, short-term bills of exchange from and to London
banks. It did not lend long-term on illiquid assets, and it had no overseas operations.
Crucially for the purposes of this study, it did not lend for the purposes of trade.
Overend’s business had been built over decades by earlier generations of partners
such that by the mid-19th century, it was called the “Corner House” in London. In the
early 1860s, a younger generation of partners took over the firm and let it be run by “wily
sycophants” who mismanaged the firm’s assets with speculative and illiquid investments
that quickly began to fail (King, 1936, p. 246).21
However, the true state of affairs was not known to the public, and the firm successfully
converted its ownership structure from a privately held company to a publicly-listed
joint-stock firm in July 1865 as a gamble to recover its losses.22 Banker’s Magazine, a
leading financial market publication, fully endorsed the firm as one of the best in the City
of London when Overend & Gurney announced its share offering. Less than one year
later, Overend announced its bankruptcy on the morning of May 11, 1866, and The Times
reported the following:
It cannot be denied that about mid-day the tumult became a rout. The doors
of the most respectable Banking Houses were besieged [...] and throngs
heaving and tumbling about Lombard Street made that narrow thoroughfare
impassable.
Overend’s failure had two immediate effects on the London money market: the first
was a negative supply shock for cash because a major intermediary could no longer fulfill
the liquidity needs of banks in London. The second was an intense positive demand
shock for bank funds as the news caused depositors to panic and run on the banks. In
conjunction, the failure froze the short-term credit market in London for several days, and
liquidity was unattainable except at the Bank of England Discount Window.23 During the
21 Appendix A.5.2 discusses the details of Overend’s business in the period before its failure.
22 Appendix A.5.2 gives the full text of the original prospectus.
23 Appendix Figure A20 plots the full time series of the daily lending at the Bank of England Discount
15
week, all London banks suffered runs, and ultimately 22 institutions were forced to close
or suspend operations. See Appendix A.5.1 for the full institutional details.
1.3 Measuring exposure to London’s crisis around the world
The goal of my empirical analysis is to estimate the causal relationship between a

location’s access to bank credit and exporting activity. I follow the literature and model
the underlying relationship between bank credit and economic outcomes by relating the
natural log of exports EXlt at location l in time t to the natural log of the amount of bank
credit:
ln( EXlt ) = a + g ln(Creditlt ) + G0 Xlt + # lt (1.1)
Identifying g from Equation 1.1 is challenging for two reasons. First, direct measures
of bank credit are of an equilibrium outcome that conflates supply and demand for credit,
so places that demand less bank credit are also likely to have less trade. Equation 1.1
will therefore not satisfy the orthogonality conditions that E[Creditlt # lt ] = 0 because # lt
includes the unobserved local economic conditions that are positively correlated with
bank credit, which biases g upward. Second, there might be reverse causality: firms in
locations that are already less productive can weaken their banks’ balance sheets through
non-performing loans and cause those banks to contract their lending or even to fail.
I overcome these two challenges by using the multinational structure of British banking
where subsidiary operations depended directly on their headquarter’s health. Banks
whose headquarters in London failed generates plausibly exogenous variation for their
subsidiary cities’ and countries’ exposure to bank failures. In the rest of this section I
describe how I measure location-level exposure to bank failures and discuss the evidence
for the identifying framework.
Window. May 11 is marked by the red vertical line. There are no extant records of Overend & Gurney’s
day-to-day operations before the crisis, so it is not clear whether depositors were acting on information
tying banks to the firm. I discuss the possible sources of information available in Appendix A.5.1.
16
1.3.1 Measuring the shock to bank credit
The total bank credit in Equation 1.1 is the sum of the credit extended by each bank b:
Creditlt = Âb Creditlbt . This location-level total can be rewritten as the sum of the shares
of each bank in a location (city or country) and the bank size: Creditlt = Âb zlbt ⇥ Creditbt
where
Creditlbt
zlbt = (1.2)
Creditlt
I calculate location l’s pre-crisis dependence (at t = pre) using the loans that were
originated in the six months before May 1866 to avoid the endogeneity of post-crisis
sorting among bad banks and bad locations. The shares zlb,pre sum to equal one in each
location.
The crisis in London generates bank-level shocks that affect locations through their
pre-crisis dependence zlb,pre on each bank. I use the shock of bank failure in 1866, which
is captured by the binary variable I (Failureb ) and takes the value of 1 if the bank failed
and 0 otherwise. Each location’s exposure to bank failure Faill is the average of failure
rates across its banks, weighted by the pre-crisis importance of each bank to a location:
Faill = Â zlb,pre ⇥ I(Failureb ) (1.3)

b
Faill takes the form of a Bartik instrument with the following first stage relationship:
D ln(Creditlt ) = a1 + b 1 Faill + G10 Xlt + nlt (1.4)
Faill is a location-level analogue to the firm-level exposures to bank-level shocks used

elsewhere in the literature, for instance in Chodorow-Reich (2014); Paravisini et al. (2014).
As in other Bartik instruments, the intuition for identification is that each location is a
small contributor to a bank’s overall operations and is therefore unlikely to drive the
bank-level outcomes. I discuss instrument validity in section 1.3.2.
17
Estimating the first stage relationship in Equation 1.4 requires location-level lending in
both the pre- and post-crisis periods. Data limitations (discussed in more detail in section
1.4.1) prevent this, but there is a strong pseudo first-stage relationship between exposure
to bank failures and credit contractions at the bank-level, shown in Table A1. Given the
lack of a first stage, the empirical results will be presented in terms of the reduced form
relationship between exposure to bank failures and the change in log exports instead:
D ln( EXlt ) = a2 + b 2 Faillt + G20 Xlt + elt (1.5)
The reduced form coefficient b 2 in Equation 1.5 is straightforward to interpret as the

semi-elasticity of the response of trade activity to bank failures in location l.24
In all calculations of the total trade credit in a location, I only observe the amounts
extended by British banks, which leads to measurement error in the endogenous variable
Creditlt . However, the instrument constructed from the shares and failure rates of British
banks will still be valid for the change in all credit as long as either non-British banks
do not provide trade credit, or the post-crisis credit supply of non-British bank credit is
uncorrelated with the failure rates of British banks across locations. Since British banks
conservatively provided over 90% of trade credit and I find no evidence of an immediate
correlation between the non-British bank response and British bank failures, it is unlikely
that the measurement error in the endogenous variable drives the results. Appendix A.3
gives the proof and empirical evidence.
1.3.2 Validity of reduced form estimation
The reduced form relationship in Equation 1.5 will causally identify the effect of con-
tractions in bank credit on exports if Faill satisfies the standard exclusion restriction for
24 Estimating the reduced form relationship means it is not possible to distinguish between the many
different roles of banking activity, such as credit provision or risk assessment. Given these banks’ role as
providers of trade credit, I focus on the credit channel, but any form of banking activity that matters for
exporters would also be affected by the bank failures.
18
an instrumental variable: E[Faill # l ] = E[Âb zlb I (Failureb )# l ] = 0. It is apparent from the
exclusion restriction that in a shift-share setting, the instrument is immediately satisfied
if bank failures are randomly assigned, but it does not require it.
The instrument will be valid if the bank-level shocks are uncorrelated with the average
location-level characteristics that determine exporting activity in the locations most
exposed to each bank (Borusyak et al., 2018).25 The identifying assumption is that banks
did not sort to locations such that characteristics of the locations were correlated with
both failures of the British multinational banks operating there and declines in exports
activity. One example of problematic sorting would be that banks that failed chose to
operate in locations that experienced a boom in the pre-period and a bust post-1866.
Declines in exports and failures of the banks operating in those locations would coincide
and be falsely attributed to the London crisis. To the extent that any boom and bust cycle
is observable, they can be included as controls in the reduced form estimation.
In the following subsections, I first show that bank failure rates themselves were not
correlated with observable characteristics of bank activity gleaned from balance sheets
nor with geographic concentration. Randomness in bank failures is sufficient to meet the
requirements for identification, but I do not rely solely on it. Next, I test the identifying
assumption directly and show that bank failure rates were also mostly not correlated with
observable characteristics of the locations where they were operating. To the extent that
certain characteristics were correlated with bank failures, they are included as controls
in all the specifications to residualize their effect on exports activity. Using the Oster
(2017) bounds, I argue that it is unlikely that there were correlations in unobserved
characteristics that would affect the results.
25 InGoldsmith-Pinkham et al. (2018), identification can come from exogeneity in the shares zlb,pre without
any information from the shocks. This condition would be satisfied if there were no sorting between banks
and locations in ways that matter for exports activity–in other words, that banks chose locations randomly
on those dimensions. However, it is likely that certain banks specialized in certain areas or commodities,
and therefore those assumptions are less suitable for this context.
19
Correlation between bank characteristics and bank failure rates
Banks are balanced across almost all observable pre-crisis bank characteristics (Table 1.1).
Panel A only has publicly-held—a.k.a. joint-stock—banks that published balance sheets,
and Panel B has all banks including privately owned banks that did not publish balance
sheets.
The balance sheet characteristics of the banks that failed are not statistically or econom-
ically different from those of the banks that did not fail (Panel A). These characteristics
are proxies for measures of bank health and risk-taking. Banks had on average £1.48
million equity capital, of which almost half was already paid by investors, and their
reserve funds, deposit liabilities, total size of the balance sheet, leverage ratio, and reserve
ratio were also similar.
In Panel B, I include all other observable characteristics that are available for all the
banks. Panel B shows that banks that survived were on average older. Age would
be a potential confounder if older banks operated in locations that were less likely to
experience declines in exports. However, the relationship is driven by private bank
outliers such as Coutts which dates from the 16th century, and the difference disappears
when those outliers are removed. Additionally, I control for the average weighted age of
banks in each location, which residualizes the age effect from the correlation between
bank credit and exports activity and leaves the residual relationship between bank credit
and exports activity.
Geographical region of specialization also did not predict bank failure. For each bank,
I calculate the total credit extended to each geographic region such as North America or
the UK itself to test whether exposure to these regions are correlated with failure.26 Banks
that failed were not more exposed to individual regions than banks that did not fail. This
balance helps to address the concern that bank failures and export contractions were
26 InTable A2, I calculate each bank’s geographic exposure as the share of total assets to rescale by bank
size. All measures are balanced there as well.
20
Table 1.1: Pre-crisis comparison of bank characteristics
Panel A: Balance sheet characteristics (joint-stock banks)
All Not Failed Failed Diff
Capital, authorized (£m) 1.48 (1.06) 1.44 (1.06) 1.67 (1.07) -0.23 (0.29)
Capital, paid up (£m) 0.59 (0.38) 0.61 (0.38) 0.47 (0.39) 0.15 (0.10)
Deposits (£m) 2.22 (2.73) 2.29 (2.82) 1.85 (2.37) 0.44 (1.14)
Reserve fund (£m) 0.13 (0.12) 0.13 (0.11) 0.15 (0.16) -0.02 (0.04)
Total size (£m) 4.81 (6.11) 5.08 (6.46) 3.73 (4.48) 1.35 (1.83)
Leverage ratio 0.24 (0.14) 0.25 (0.14) 0.23 (0.11) 0.02 (0.05)
Reserve ratio 0.06 (0.07) 0.06 (0.07) 0.06 (0.06) 0.01 (0.03)
N 95 76 19 95
Panel B: Other characteristics (all banks)
Trade credit (£k) 105.79 (246.77) 112.57 (264.53) 73.16 (130.51) 39.41 (57.9)
Age (years) 35.91 (53.62) 40.88 (57.16) 11.33 (15.37) 29.54 (12.6)**
Cities (#) 13.75 (22.83) 14.90 (24.56) 8.23 (9.80) 6.67 (5.3)
Countries (#) 7.62 (8.89) 7.90 (9.26) 6.32 (6.84) 1.58 (2.1)
Asia (£k) 46.04 (170.08) 49.42 (184.96) 29.74 (59.65) 19.68 (40.0)
Africa (£k) 8.17 (25.08) 7.13 (21.95) 13.20 (36.90) -6.07 (5.9)
N. America (£k) 13.59 (44.91) 15.65 (48.79) 3.68 (13.07) 11.97 (10.5)
S. America (£k) 6.99 (34.12) 7.79 (37.25) 3.13 (9.21) 4.66 (8.0)
Australia (£k) 6.41 (17.25) 7.00 (18.58) 3.58 (7.87) 3.42 (4.0)
Europe (£k) 12.21 (27.39) 10.87 (25.41) 18.70 (35.41) -7.83 (6.4)
Brit. Emp. (£k) 48.25 (149.40) 53.47 (162.52) 23.13 (46.02) 30.34 (35.0)
UK (£k) 12.37 (39.96) 14.70 (43.56) 1.14 (2.67) 13.57 (9.3)
N 128 106 22 128
Notes: Table 1.1 Panels A and B shows bank-level balance across characteristics for banks that failed and
did not fail. All variables are measured at the end of 1865 before the crisis. Balance sheet variables were
only published for publicly traded banks; these are reported separately in Panel A. "Not Failed" and
"Failed" refers to whether a bank suspended or closed during the crisis. Means are reported first, and
standard deviations are given in parentheses. "Diff" refers to the difference in means between groups.
Standard errors are reported in parentheses for the "Diff" column. £k denotes units of thousands of pounds
sterling. £m denotes units of millions of pounds sterling. Leverage ratio is defined as capital (paid and
reserves) divided by total assets. Reserve ratio is defined as reserve assets divided by deposit liabilities.
Significance is marked by *p < 0.1, **p < 0.05, ***p < 0.01. Sources: Bank of England Archives C24/1,
Banker’s Magazine, The Economist.
21
simultaneously caused by a shock that was systematically correlated with their geography.
Examples of such shocks include weather patterns that led to widespread crop failures
and declines in output or regional boom-and-bust patterns. In addition, banks in the two
groups were similarly geographically diversified, operating in an average of almost 14
cities and 8 cities.
Correlation between location characteristics and bank failure rates
Bank headquarters were exposed to shocks in London, but these shocks could be corre-
lated with the characteristics of the banks’ subsidiary locations. Correlations between
location-level characteristics and a location’s exposure to bank failures are problematic if
those characteristics are the ultimate drivers of exports activity. For example, if the banks
that failed were primarily operating in countries focused on cotton production, and those
countries were also the ones with the largest declines in exports, then shocks to the cotton
industry could simultaneously be causing both the bank failures and exports outcomes.
One way to test the exogeneity of bank-level failure rates to location-level characteris-
tics is to calculate each bank’s exposure to those characteristics and correlate them with
the bank failure rates (Borusyak et al., 2018). The advantage of testing the bank-level
relationship rather than the location-level relationship, the latter of which is standard
in the literature, is that performs the standard error correction described in Adão et al.
(2018).27
I examine the observable pre-crisis location-level characteristics at both the port-level
and the country-level, since those are the two units of observation I use. At the port-level,
the observable characteristics include the volume of exports (proxied by the number of
ships from the Lloyd’s List), the importance of the United Kingdom as a destination, the
27 Adão et al. (2018) show that when the source of identification from a shift-share instrument are the
shocks, the standard errors of regressions of the instrument on location characteristics tend to over-reject the
null hypothesis. Intuitively, the location-level tests target randomness in the shares, but when the location
shares themselves are not suitable instruments, the covariance between the shocks and the shares may be
relevant. Borusyak et al. (2018) show that implementing the Adão et al. (2018) standard error correction is
equivalent to translating the location-level characteristics into bank-level exposure rates.
22
geodesic distance to London, the latitude, the number of destinations, the availability of
non-British banks, and whether the port is a capital city.28 At the country-level, observable
characteristics include the total value of exports, the value of exports within industries,
the share of commodities in the composition of exports, the currency system, and whether
the country was engaged in conflict. These characteristics help to capture heterogeneity
in size and trade patterns. Each bank’s share-weighted average exposure X̄b to these
Âl zlb ⇥ Xl
pre-crisis characteristic Xl is calculated as X̄b = Âl zlb
where larger weights are given
to locations more dependent on bank b. The transformed location-level characteristics X̄b
are normalized and individually regressed on bank failure rates:29
X̄b = a + bI (Failureb ) + # b (1.6)
Table 2.3 reports the results and shows that there is balance on almost all character-
istics.30 While most observable characteristics are uncorrelated with failure rates, it is
still possible that other unobservable characteristics are correlated. In the main empir-
ical analysis, I rely on the Oster (2017) bounds to argue that the degree of unobserved
heterogeneity would have to be unreasonably large to drive the main results.
In terms of port-level characteristics, Panel A shows that two factors are unbalanced:
banks operating in ports with a higher fraction of exports going to the UK were more
likely to fail, and those operating in ports that were also the capital cities within countries
were less likely to fail.31 These characteristics will all be included as controls in the
28 Results are similar using sailing distance (without access to the Suez Canal) instead of geodesic distance
to London. Figure A5 plots the relationship between the two types of distances and discusses the data
sources.
29 The regressions are weighted by ẑ , which is the average location exposure to bank b: ẑ = 1 L
b b L Âl =1 zlb .
The weighting is necessary to translate location-level relationships to bank-level relationships. The full
derivation for the equivalence is given in Borusyak et al. (2018).
30 It is worth noting that given the number of hypothesis tests being run, it would not be surprising for
some of them to reject the null.

31 As discussed in Borusyak et al. (2018), the advantage of transforming all the specifications into shock-
level (bank-level) regressions is that it makes it clear which shocks (banks) are the most relevant for the
results. At the port-level, port cities are matched to the closest geographic city of financing, which makes
it possible that some cities are not the closest for any port. If certain banks operated in only unmatched
23
Table 1.2: Correlation between bank failures and pre-crisis location characteristics
X̄b = a + bI (Failureb ) + # b
Panel A: Port characteristics
Ships Ships stm Frac to UK Dist to London Latitude Non-Brit banks Destinations Capital city
(1) (2) (3) (4) (5) (6) (7) (8)
I(Failure) 0.197 0.127 1.032*** -0.161 0.362 -0.433 -0.399 -0.666***
[0.227] [0.246] [0.213] [0.164] [0.212] [0.313] [0.250] [0.201]
N 122 122 122 122 122 122 122 122
Panel B: Country characteristics

Exports values Frac commodities Gold Silver Bimetallic Conflict: any Conflict: interstate
(1) (2) (3) (4) (5) (6) (7) (8)
I(Failure) 0.239 -0.221 -0.404 -0.00674 0.344 0.208 -0.0530 0.346***
[0.130] [0.261] [0.216] [0.192] [0.215] [0.178] [0.201] [0.0894]
N 128 128 128 128 128 128 128 128
Panel C: Country characteristics: exports composition
Cotton, raw Cotton, manu Grains Bullion Sugar Coffee Alcohol Tobacco
(1) (2) (3) (4) (5) (6) (7) (8)
I(Failure) -0.0375 -0.0925 0.106 -0.0457 0.384** -0.0622 -0.146 -0.0608
[0.109] [0.0672] [0.0727] [0.0730] [0.167] [0.197] [0.216] [0.0623]
N 128 128 128 128 128 128 128 128
Notes: Table 2.3 reports estimates from the bank-level regression of bank exposure to location
characteristics pre-crisis on bank failure rates. The dependent variable is X̄b , the share-weighted exposure
of banks to location characteristics, normalized to have zero mean and unit variance. The coefficients are
interpreted as the standard deviation increase in the average bank exposure to a particular characteristic if
the bank failed. Panel A includes location characteristics from the port panel. There are 122 observations
instead of the full 128 because 6 banks operated in cities which were not the closest city for any port.
Panels B and C includes country-level characteristics like the monetary standard and presence of conflict in
the exporting country in 1865/1866, and the industry composition of exports in 1865. Regressions are
weighted by each the average location’s exposure to bank b. *p < 0.1, **p < 0.05, ***p < 0.01
24
baseline specifications to residualize the direct effect that they have on any decline in
exports.
In order to address the possibility of commodity booms and busts, I categorize each
country’s exports by two-digit SITC categories and test balance across the top eight
categories. The full distribution of exports by SITC categories is plotted in Figure A1.
Raw cotton and cotton manufactured goods are the largest components of textile fibers
(category 26) and textiles (category 65), respectively, but I isolate these from their two-
digit categories because of their historical significance. In particular, in 1866 after the
American Civil War ended, there was a large disruption in global cotton markets as the
US South began producing cotton again (Beckert, 2015). Banks exposed to the post-war
cotton shock, either because they specialized in the cotton trade or because they operated
in cotton-exporting countries, could have failed because of disruptions to the cotton
market and exports from those places could have fallen for the same reasons, leading to a
spurious correlation between bank failures and declines in exports.
Table 2.3 Panel B shows that there is no correlation between exposure to different
currency standards (gold, silver, or bimetallic) and bank failure rates. Panel B also checks
for balance in exposure to conflicts with interstate conflicts separated from all other types
(intrastate and extrastate). There is a strong correlation between exposure to non-interstate
conflicts and bank failures, but these effects are driven by the small number of those
types of conflicts. Table 2.3 Panel C shows that banks that failed were not differentially
exposed to either raw cotton exports or cotton manufactured goods. There is also balance
across the other major commodities, including bullion, grains, coffee, alcohol, and tobacco.
However, banks operating in countries that exported more sugar were more likely to fail.
The location-level characteristics that are correlated with bank failure rates are included
as controls in the main empirical specifications to address their potentially confounding
cities, they would be irrelevant in the port-level relationship between exposure to financing and declines
in exports. The smaller number of observations in Panel A reflects exactly this fact: at the port-level, six
banks operated in cities that were not matched to any ports. These are smaller banks, and excluding them
entirely makes no difference at the country-level.
25
effects.
1.4 Data
This paper combines several newly collected and digitized historical datasets. In this
section, I give an overview of the most important datasets and variables that I constructed.
I provide full details, discussion, and documentation in Appendix A.6.
1.4.1 Exposure to bank failures
I use the Bank of England’s handwritten records of city-level lending by banks pre-crisis
to calculate the importance of banks to locations, zlb,pre . The Bank of England kept
detailed records of every transaction that occurred at its Discount Window. Banks facing
their depositors’ demands during the banking crisis discounted bills of exchange at the
Bank of England because it was the only source of liquidity during the crisis. I interpret
the bills that these banks brought in for discount as an unbiased representation of the
universe of loans extended by British banks in locations around the world. One concern
is that the bills discounted by the Bank of England suffer from selection bias because
worse banks may have held worse collateral, and the bills they held are underrepresented
in the data. However, the relevant selection is at the bank-location level, not just at the
bank-level. In order for this type of selection to be driving the results, it would need to
be the case that locations with export growth are attributed with falsely low measures
of exposure to bank failures, and vice versa for locations with export contractions. All
contemporary and modern evidence on the London money market indicate that by the
mid-19th century, the only relevant attributes of bills were the banks that accepted them
and their maturity. Several additional institutional details provide evidence that selection
is unlikely to be an issue. I discuss these in detail in Appendix A.5.3.
I use the ledgers from 1865-1866 to build a dataset of over 11,000 individual loans from
26
the 128 banks that had international operations in the year before the crisis. An example
of a ledger page is shown in Figure 1.1a. For each handwritten loan record, I document
the bank that originated and guaranteed the loan, the city the loan was extended in, the
amount of the loan, and the date it was brought to the Bank of England to be discounted.
Deciphering the hand-writing was not trivial. When there was uncertainty about the
city of origination, I looked for other loans extended to the same borrower to compare
entries. I was able to identify the location and geocode 99.7% of the value of loans. These
banks operated in a total of 180 cities outside of the United Kingdom, and they lent
over £11.2 million in the year before the crisis. The general lack of data on lending via
bills of exchange has been well-documented, and to my knowledge, there are no other
comprehensive empirical studies of British bank-intermediated finance during this period
despite their role in global financial markets.32
Figure 1.2a maps the geographic distribution of exposure to bank failures, Faill at
the city level. The size of the points measures the pre-crisis amount of British lending
in the city, and the color portrays the bank failure share. This map shows within and
across-country variation in failure rates. Figure A2 plots the full distribution of exposure
across ports and countries.
1.4.2 Immediate outcomes
I build a port-level panel of bilateral shipping activity for ports outside the United
Kingdom using the daily publications of the Lloyd’s List newspaper for the years 1865-
1867. An example of this source from September 5, 1866 is shown in Figure 1.1b and
32 Scholars have attempted to estimate the aggregate size of the trade bills market with the “stamp revenue”
(taxes), but these are poor estimates and contain no geographic detail (Nishimura, 1971a). Nishimura
(1971a) notes that the other source of records would be the surviving balance sheets from a few of the
largest banks during the period, but they similarly have no geographic detail. Jones (1995) estimates the
geographical distribution of total assets of British multinational banks for certain benchmark years between
1860-1970, but he does so by defining broad regions (such as Asia, North America, Europe without the UK)
and attributing all of a bank’s assets to that region. These data are informative of broad patterns, but they
are too limited for empirical studies. Reber (1979) discusses the general lack of records that survive from
the international subsidiaries of British banks.
27
Figure 1.1: Data sources
(a) Excerpt of the Bank of England Discount Office ledgers
(b) Excerpt of the Lloyd’s List
Notes: Data for Figure 1.1a come from Bank of England Archives C24/1. This is an example of the original
records used to construct the financing data. The name of the bank, Agra and Masterman’s, is written at
the top. The column on the far left, “Whence Drawn,” give the city where the credit was originally issued.
The column on the far right, “Upon,” gives the values of the loans. Data for Figure 1.1b come from the
British Library. This excerpt from the Lloyd’s List of September 5, 1866 show the organization of the records
and the typical information available. Under each port, ships are listed individually with their name, their
captain’s name, type of ship, whether they arrived to the port or sailed from it, the destination of their
movements, and the date of the event. Coastal (i.e. domestic) trade was omitted from the records for
non-British ports.
28
Figure 1.2: Geography of banking and trade
(a) British multinational bank lending and failures
(b) Port-level trade activity
Notes: Figure 1.2a maps the distribution of the city-level exposure to bank failures Faill . The size of the
points denote the log value of total credit at each city and the color gradient denotes the exposure to
failure, ranging from 0 to 1. Figure 1.2b maps the distribution of shipping activity at ports in the pre-crisis
year. The size of the points denote the log number of ships leaving. Ports in the United Kingdom are not
included. Source: Lloyd’s List.
29
the method for extracting the route info is described in appendix A.6. I digitized the
daily newspapers for all shipping events and geocoded 99.8% of the origination ports
to 377 unique ports. Over 8,000 unique destinations were geo-coded and assigned to 60
countries.33 Figure 1.2b maps the distribution of pre-crisis activity levels for the ports
around the world where the size of the dots denotes the log number of ships. One
drawback of the Lloyd’s List data is that it does not report values of the goods onboard.
However, there is a strong positive correlation between the number of ships leaving a
country in a year and the total value of the country’s exports, shown in Figure A3.
1.4.3 Long-term outcomes
For the long-term outcomes, I measure exports and access to bank-intermediated finance
for the period 1850-1914. The country-level panel of bilateral trade are constructed from
publicly available datasets of historical trade statistics along with my own contributions to
create a meta-dataset that is, to my knowledge, the most comprehensive available. These
datasets cover a variety of time periods and territorial border changes, so I standardize
country definitions to the smallest landmass unit that is consistently reported over all the
years.34
I measured access to bank-intermediated finance at the city-level in five-year intervals
by digitizing the annual editions of the Banking Almanac. I assigned the banks nationalities
according to the Banking Almanac when available and other primary sources. Table 1.3
reports the descriptive statistics for ports and countries in 1865. The average port saw
130 ships leaving in the pre-crisis period and had 7 pp exposure to failed banks with one
standard deviation of 19 pp. The average country-level exposure to bank failures was 11
pp with a standard deviation of 17 pp.
33 Destinations are inconsistently listed as countries or cities, so they are aggregated to a larger unit of
observation. This also minimizes sparsity in the dataset while remaining an effective way to control for
demand-side shocks.
34 These units most closely resemble pre-WWI borders.
30
Table 1.3: Summary statistics: Ports and countries
Ports Countries
mean median sd mean median sd
Exposure to failed British banks 0.07 0.00 (0.19) 0.11 0.03 (0.17)
Exposure in British Empire 0.03 0.00 (0.05) 0.08 0.06 (0.10)
Exposure outside British Empire 0.10 0.00 (0.22) 0.12 0.02 (0.20)
Exports 127.99 32.00 (231.05) 12.49 2.15 (32.96)
Fraction exports to UK 0.39 0.30 (0.34) 0.62 0.69 (0.37)
Destinations (# countries) 7.60 5.00 (7.28) 3.95 2.00 (8.32)
Distance to destination (’000 km) 5.31 5.12 (3.48) 6.12 5.26 (3.51)
Banks 6.03 3.00 (7.54) 5.27 1.00 (9.96)
Non-British banks 0.60 0.00 (1.06) 2.97 0.00 (8.74)
Fraction in British Empire 0.34 0.00 (0.47) 0.33 0.00 (0.47)
N 289 55
Notes: Table 1.3 shows summary statistics from the port-level panel of shipping activity and the
country-level panel of values of exports. All variables are measured at the end of 1865, before the crisis.
"Exports" is measured by the number of ships departing for ports, and by the value of exports in millions
of pounds sterling for countries. Fraction of exports to the UK is similarly calculated using the number of
ships and values of exports.
31
1.5 Immediate impact on trade
This section contains my results on the immediate effect of bank failures on exporting
activity on both the intensive and extensive margins. There would be a contraction in
exports if bank failures raised the cost of financing sufficiently for trade to be unprofitable.
I first identify the effects using within-country variation from port-level shipping activity
before turning to across-country variation with country-level shipping activity and values
of exports.
1.5.1 Intensive margin effect: baseline specification
I examine the immediate impact of bank failures on exports using the two-period panel of
port-level shipping activity. Each port in the port-level panel is matched to the closest city
of financing by geodesic distance, and its exposure to bank failures Fail po is assumed to
come from that city. For example, the port of Piraeus in Greece is designated as receiving
its funding from Athens. This empirical strategy is based on the theoretical and empirical
evidence that banks operate locally.35 Ports more than 500 km from the nearest city of
financing are given an exposure of 0, and I include a time-varying intercept for these
ports so that there is a control group of completely unexposed ports.36 This control group
allows for ports that are still connected to London but experienced no bank failures to
react differently from ports that were not connected to London at all.
In the raw data, there is a strong negative correlation between exposure to bank failures
and the difference in the log number of ships sailed in the post-period relative to the pre-
period. Figure 1.3a plots the binscatter and linear fit within-country at the port-level, and
35 Sharpe (1990) presents a theoretical framework for why contracting frictions between banks and
borrowers are higher at greater distances. Petersen and Rajan (1994, 2002) present empirical evidence on
the importance of geographical proximity to lending activity.
36 The results are not sensitive to the 500 kilometer boundary and the main coefficients are robust for
a range of distances. The results are also robust to not including the time-varying intercept for distant
ports. See Figure A4 for the coefficient plot for the baseline specification estimated using different distance
cutoffs.
32
Figure 1.3b shows a similarly negative relationship across countries. I formally estimate
the effect of bank failure exposure on exports in a difference-in-difference regression:
ln(S pot ) = bFail po ⇥ Postt + a p + got + # pot (1.7)
where S is the number of ships leaving from port p in origin country o in period t.
Following the trade literature, the dependent variable is in logs to reduce the effect of
outliers. As in Paravisini et al. (2014), I separate the intensive and extensive margin effects
rather than transforming the zeros. The intensive margin sample is constructed from
shipping activity five quarters pre- and post- May 1866 and limited to ports active in both
periods.37
b is the coefficient of interest, which we would expect to be negative if increases in
the cost of financing from bank failures reduced exports. Postt is an indicator for the
post-crisis period that control for macroeconomic shocks affecting the exports trend over
time. For example, changes due to the overall level of interest rates following the crisis
would be absorbed this way. Port fixed effects a p absorb all time-invariant port-specific
differences in levels of shipping, including differences correlated with their exposure to
bank failures. Origin-country-period fixed effects got flexibly control for all observed
and unobserved characteristics at the country-level that affected shipping. Insofar as
ports within countries exported a similar composition of goods pre-crisis, these serve as
proxies for any country-level industry specialization shocks such as factor endowment
and factor price movements. Including these fixed effects means b is identified off within
origin-country variation in exposure to bank failures.38 Regressions are weighted by the
pre-crisis size of ports, measured by shipping activity in the pre-crisis year, to estimate
the economically meaningful average effect and to avoid confounding the estimation
37 I choose 5 quarters to estimate roughly 1 year pre- and post-crisis, allowing for lags in the response
time.
38 Countries with only one port are effectively dropped from this estimation. These account for 16 of the
578 observations (2.8 percent). These come from 8 ports, which reduces the effective number of countries in
the estimation from 54 to 46.
33
Figure 1.3: Correlation between exposure to bank failures and shipping in 2 year window
(a) Port-level relationship
.4
.2
Δ ln(ships)
-.2
-.4
0 .2 .4 .6 .8
Port-level exposure (Failpo)
(b) Country-level relationship
1.5
SWENOR
1
AUTHUN
CHL
Δ ln(Ships)
VIR BEL
JPN URY
.5
PRTPHL
GIB USAAUSNZL
PER
SLE TTO
GTM RUSPAN
ESP
DNK
THA ZAF
ARG
BRA
AZORES
VEN
SHN
MUS
LKA
CAN
COL
GBRWINDIES CUB
HKG
GRC
0 JAM
IDN
MEX
DEU
EGY
OTTO
GUY GBRIND MLT
ITA CHN
FRA
STRAITS
ROU
-.5
PRI NLD
ANT
0 .2 .4 .6 .8
Country-level exposure (Failo)
Notes: Figure 1.3a is a binscatter plot of the correlation between the change in the ln number of ships from
the post-crisis period to the pre-crisis period (for the crisis occuring on May 11, 1866) and the port-level
exposure to bank failures. This plot is residualized on country-level shipping so it plots the within-country
relationship. Figure 1.3b is a scatterplot of the correlation between the change in the ln number of ships
and country-level bank failures. Countries within the British empire are marked with a red triangle. The
full list of country abbreviations (some of which are non-standard to account for colonies) is given in
Appendix A.5.4.
34
with an endogenous post-crisis response. Standard errors are clustered by the country of
origin to allow for heteroskedasticity and within-country spatial correlations.39
The estimation strategy compares outcomes in port cities that received a large financing
cost shock to those that received a small shock before and after the London banking
crisis. Unlike a standard DD, the treatment intensity is continuous. The distribution of
treatment is well-represented across the entire range of exposure (Figure A2a).
Identification requires that there are no shocks correlated with the bank failures that
occurred simultaneously. First, I address these concerns by controlling for all location-level
characteristics that are correlated with bank failures. Second, I use another characteristic
of the historical context—the nascent international telegraph system—to show that the
timing of the effect is consistent with when the news from London would have reached
the ports.
Baseline results
Table 1.4 presents the baseline results with controls added individually. The point estimate
in column 1, estimated across all ports without the country fixed effects, indicates that
ports exposed to complete British bank failure shipped 68.7 percent less than unexposed
ports in the post-crisis year. The within-country comparison in column 2 gives a similar
magnitude. The similarity in the estimates implies that differences in origin-country
characteristics are not driving the main results.
These magnitudes are larger than those estimated by Amiti and Weinstein (2011)
and Paravisini et al. (2014), who estimate the effect of bank-level shocks on Japanese
and Peruvian firms, respectively.40 There are two likely reasons for the difference: first,
39 Clustering at the country-level is reasonable because exporting activity is likely to be more correlated
within countries than across them. Standard errors could also be clustered by the city of financing to
account for serial correlation. Results are robust to clustering by this lower level of aggregation.
40 Amiti and Weinstein (2011) proxy bank health with a decline in its market-to-book value. Scaling their
main coefficient in Table 3 to a 100 percent decline in market-to-book would imply that a firm’s annual
exports values declined by 9 percent. Paravisini et al. (2014) instrument for bank credit supply with the
bank’s dependence on foreign funding. Scaling their baseline IV coefficient in Table 5 to a 100 percent
contraction in bank credit supply would imply a 19.5 percent contraction in the volume of exports.
35
Table 1.4: Immediate effect of bank failures on port-level shipping
(1) (2) (3) (4) (5) (6) (7) (8)

Failpo ⇥ post -0.687*** -0.655*** -0.670*** -0.659*** -0.596*** -0.557*** -0.394*** -0.327**
[0.247] [0.124] [0.123] [0.203] [0.129] [0.122] [0.131] [0.133]
Capital city ⇥ post Y
Age of banks ⇥ post Y
# non-Brit banks ⇥ post Y
Fraction to UK ⇥ post Y
Destinationd ⇥ post FE Y Y
ln(distanceod ) Y
Countryo ⇥ post FE Y Y Y Y Y Y Y
Portp FE Y Y Y Y Y Y Y Y
N 578 578 578 578 578 578 2532 2532
Ports 289 289 289 289 289 289 262 262
Destinations 54 54
36
Clusters 54 54 54 54 54 54 51 51
b* -.693 -.66 -.675 -.665 -.598 -.557
d 60.98 45.46 47.68 34.96 39.17 41.11
Notes: Table 1.4 reports estimates from the difference-in-difference regressions from the two-period panel of port-level shipping activity in the year
before and after the crisis. The dependent variable in columns 1 to 6 is the ln of the total number of ships departing in each period. The dependent
variable in columns 7 and 8 is the ln of the number of ships departing for each destination in each period. Fail po is the share of the port’s banks that
failed during the crisis. The mean of Fail po is 0.07, and the standard deviation is 0.2. post is a dummy for the post-crisis year that takes the value of
1 after May 1866 and 0 otherwise. The time-invariant control variables are measured in 1865 and interacted with the post dummy. They include an
indicator for the port being a capital city within the country, the average ln age of banks, the number of non-British banks, and the fraction of
shipping to the UK. Column 8 controls for the ln geodesic distance between the origin and destination countries. The sample is restricted to ports
ship in both the pre- and post-period. Results from implementing the Oster (2017) test of selection on unobservable characteristics are reported in
the last two rows. b⇤ is a bound on Fail po ⇥ post if selection on unobservable is as large as selection on unobservables (d = 1). d is the degree of
selection on unobservables necessary for the estimated coefficient to be 0. Standard errors in brackets are clustered by country of origin. *p < 0.1,
**p < 0.05, ***p < 0.01
complete bank failure is a qualitatively more extreme outcome than declines in bank
health; second, financing frictions in the 19th century are most likely larger than in the
modern-day because information frictions were much higher.
Robustness to controls
I address the concern that the bank failures are correlated with other factors that are
responsible for the decline in exports by including observable location-level character-
istics as controls in the baseline regression. These controls are based on the port-level
characteristics that were not balanced between banks that failed and did not fail in Table
2.3 and deal with any confounding effect they may have in driving the results. They
include the number of non-British banks, the average age of the banks, whether the port
is the capital city, and the fraction of ships going to the UK in the pre-crisis year.41
The coefficients in Table 1.4 columns 3-6 after including these controls remain stable
and statistically significant. Column 7 shows the coefficients after including all con-
trols. Implementing the recommended bounds in Oster (2017) shows that selection on
location-level unobservable characteristics is minimal. These bounds are calculated using
changes in the magnitude of the coefficient and the R2 after controlling for observable
characteristics. b⇤ is the inferred true coefficient if the unobserved bias is as large as the
observed bias, and d is the inferred bias that could induce the estimated b to be zero. I
report these as b⇤ and d in the last two rows. These calculations show that b⇤ is almost
identical to the estimated b, and that the degree of unobservables bias would have to be
at least 35 times larger than the degree of observables bias.
The baseline effects are also not due to demand shocks. Since the United Kingdom
accounted for 30% of global trade during this period, a particular concern is that unob-
served declines in UK demand are driving the results. I modify Equation 1.7 so that the
dependent variable is ln(S podt ) where S podt is the number of ships sailing from port p
41 Bank-levelcharacteristics are aggregated to the port-level using the pre-crisis shares zlb,pre of the
importance of each bank to each location.
37
in country o to destination country d in period t, and I include destination time-trends
gdt .42 gdt will accommodate all import demand shocks that might be confounding the
effects, especially those from the United Kingdom. In this specification, b is estimated off
the variation across ports shipping to the same destination-country.43 As before, I limit
the sample to origin-destination pairs that ship in both periods to isolate the intensive
margin effect. Table 1.4 column 7 reports a coefficient of -0.39, which is smaller than the
baseline coefficient, but statistically significant at the 1 percent level as before.
Although there is a large amount of heterogeneity in the treatment, the binscatter in
Figure 1.3a might raise concerns that the results are driven by a few outliers. I show that
this is not the case. Results are robust to trimming or winsorizing the top and bottom 10
percent of the observations.
Allowing for news lags
So far, I have assigned a single treatment date for all ports in the DD estimation. However,
in reality there were long communication lags in the mid-19th century because the global
telegraph network was not fully connected. Basing the post-crisis event date on May 11
for all ports around the world falsely attributes pre-crisis shipping events to the post-
crisis period for ports far away from London, which can bias the difference-in-difference
estimates. An alternative method bases the event date of the crisis for each port on the
date that news from London would have reached the port. For all ports, I calculate the
average news lag between when shipping events occurred and when it was reported
in the Lloyd’s List.44 For major cities, I validate these calculations with the first local
newspaper reporting of the banking crisis.
42
Â p S podt = S pot : the sum of shipping to all destinations is equal to the dependent variable in the
baseline specification.
43 Destinations that only ship from single ports within origin countries are effectively dropped from the
estimations. These singleton observations account for 5 of the 2,532 observations.

44 Juhász and Steinwender (2017) similarly use lags in the Lloyd’s List reports to measure communication
times to London before and after the global telegraph network was established. Juhász (2018) uses the
Lloyd’s List data to track port activity during the Napoleonic blockade and document its reliability as a
source for trade flows.
38
Communication times are highly correlated with the geodesic distance, although there
are outliers due to the burgeoning telegraph network. Figure A6 shows the relationship
between (geodesic) distance to London and the average news lag in days. The last cities to
receive the news were those in the interior of China and New Zealand. To allow for some
flexibility in the effective arrival date, I mark the month of the news date as spanning
two weeks on either side of the calculated news arrival date. I build a balanced panel
of shipping activity around the news arrival date to that port. I validate the port-level
results using the port-specific news arrival dates to mark the post-period and report the
estimates in the appendix.
Intensive margin effects: values of trade
Shipping S is a proxy for the volume of exports which may overstate the true effect
if there was an increase in the capacity utilized on ships post-crisis; conversely, it will
understate the true effect if ships were filled to lower capacity post-crisis. In addition,
overland trade will not be captured by ship movements. I overcome the limitations in
the Lloyd’s List shipping data by using the annual country-level bilateral values of trade
dataset to estimate the effects of the bank failures over calendar years.
I estimate the short-term losses in a dynamic difference-in-differences specification for
the years 1865-1870:
ln(EXodt ) = b t Failo + G0 Xot + go + gdt + qt ln(dist)od + # odt (1.8)
This specification includes leads and lags to the shock interacted with treatment
Failo which makes it possible to visualize any pretrends and the evolution of the effect
over time. The dependent variable is the log value of exports EXodt (in nominal pounds
sterling) from origin country o to destination country d in year t. b t is the coefficient of
interest, which is estimated every year and captures the semi-elasticity of exports values
39
from country o to country d to bank failure exposure.
As in the port-level estimation, I control for the effect of the origin country not having
any British banks at all in 1866, which separates the effect of any exposure from the
degree of exposure to failed banks.45 Xot includes pre-crisis country characteristics that
are interacted with a post-crisis dummy. Destination-country year fixed effects gdt control
for demand shocks to address the concern that countries exposed to bank failures were
exporting to destinations that contracted their demand for other reasons.46 I omit the
covariate for the first year at t = 1865 in the estimation and normalize it to zero. Standard
errors are clustered at the unit of treatment, the exporter country, following Abadie et al.
(2017).47
Equation 1.8 is the fixed effects estimation of a structural gravity model standard in the
international trade literature (Head and Mayer, 2014). Gravity models relate the volume
of trade flows to the sizes of the importing and exporting countries and the inverse of
the distance (geographic and institutional) between them.48 I control for the distance
between countries distod as a standard measure of bilateral resistance. Allowing qt to
vary by year flexibly controls for shocks to the effective distance between countries due
to technological advances. The one departure from the standard fixed effects estimation
using panel data is the absence of origin-country year fixed effects because those are
collinear with the treatment. However, to the extent that Failo affects other economic
conditions (such as GDP) that also affect exports, origin-country year characteristics are
an endogenous outcome and not a suitable control.49
Table 1.5 presents the results for the coefficient on Failo estimated annually. The
45 These countries accounted for 2% of the value of exports in 1866, and results are robust to not controlling
for the non-exposed group.
46 Including g as a control variable restricts the estimation to destination countries that import from
dt
more than one country.
47 Other work has concluded that it is important to account for the dyadic nature of trade data (Cameron
and Miller, 2014). I show that results are robust to different ways of clustering in Table A5.
48 Gravity can be micro-founded from most international trade models, including ones featuring perfect
competition, monopolistic competition, and monopolistic competition with fixed costs of entry.
49 Most applications of gravity study the effect of bilateral trade shocks, such as a regional trade agreement
or a currency union, which allows for the shock to be at the bilateral country level.
40
coefficient b 1865 is statistically indistinguishable from 0 across all specifications, which
confirms that there were no pre-trends in the outcomes, and that the decline in trade
was concurrent with the banking crisis. In column 2, b 1867 is interpreted as the log-point
decline in exports in 1867 relative to 1866 in countries exposed to bank failures relative to
countries not exposed, all exporting to the same destination country. The magnitude is
around -1 in all specifications meaning that the average country (exposed to 11 percent
bank failures) exported 9.1 percent less in the year after the crisis. This coefficient is larger
than the baseline from Table 1.4, which suggests that ships were likely filled to lower
capacity post-crisis. Using the longer panel of outcomes also shows that the contractions
in 1867 worsen in 1868 and are economically and statistically lower every year until
1870. In section 1.6.1, I explore the long-term effects until the end of the First Age of
Globalization in 1914.
1.5.2 Extensive margin effects
Many models of international trade have firms paying a fixed cost in order to export their
products (Chaney, 2016; Melitz, 2003). In these models, shocks to the cost of capital will
impact the extensive margin of exporting activity if exporters use external finance to pay
fixed costs of entry. Empirically, the extensive margin of entry and exit into exporting
activity has been shown to explain a large share of the variation in trade flows (Helpman
et al., 2008).
I categorize the extensive margin of exporting activity in two ways: the first is the
number of unique destinations that a port trades with conditional on trading at all, and
the second is the likelihood that a port engages in any international trade. I estimate
extensive margin losses to the number of destinations using the specification in Equation
1.7 with the log of the number of unique destinations as the dependent variable. I report
the within-country results in Table 1.6 column 2: ports completely exposed to bank
failures exported to 29.5 percent fewer destinations than unexposed ports. The effects are
41
Table 1.5: Immediate effect of bank failures on country-level values of exports
ln(EXodt ) = b t Failo + G0 Xot + go + gdt + qt ln(dist)od + # odt

(1) (2) (3) (4) (5) (6)
b1865 -0.208 -0.240 0.0690 -0.183 -0.260 -0.236
[0.198] [0.214] [0.155] [0.229] [0.314] [0.216]
b1867 -0.842* -0.921 -1.038 -0.920 -0.921 -0.999
[0.446] [0.603] [0.647] [0.603] [0.603] [0.643]
b1868 -1.835*** -1.611*** -1.732** -1.611*** -1.612*** -1.599***
[0.410] [0.551] [0.769] [0.551] [0.551] [0.568]
b1869 -1.883*** -1.872*** -1.844*** -1.871*** -1.872*** -1.931***
[0.338] [0.410] [0.447] [0.409] [0.409] [0.418]
b1870 -1.669*** -1.633*** -1.389*** -1.632*** -1.633*** -1.607***
[0.349] [0.434] [0.434] [0.434] [0.433] [0.443]
Controls Y Y Y Y Y Y
Countryo FE Y Y Y Y Y Y
Regionot FE Y
ln(cottono ) ⇥ Post Y
ln(cotton manuo ) ⇥ Post Y
ln(populationo ) ⇥ Post Y
I(Brit bankot ) Y Y Y Y Y Y
Countryd Y
Countrydt Y Y Y Y Y
N 2952 2952 2952 2952 2952 2571
Clusters 83 83 83 83 83 67
Adj. R2 0.573 0.551 0.543 0.551 0.551 0.546
Notes: Table 1.5 reports estimates from the annual dynamic difference-in-difference regressions from the
panel of country-level values of trade. The dependent variable is the ln value of exports from origin
country o to destination country d. There are 83 exporting countries from 1865-1870. Failo is the share of
the country’s banks that failed. post is a dummy for the post-crisis years 1867-1870. Baseline controls are
the log distance between country o and country d. Cotton, cotton manufactured goods, and population are
calculated in 1865 and interacted with the post dummy. Countries that did not export cotton are given ln
values of zero. Controls for the log of population reduces the sample size due to data limitations. Standard
errors in brackets are clustered by the origin country. *p < 0.1, **p < 0.05, ***p < 0.01
42
even stronger at the country-level. These results provide suggestive evidence that there
were negative spillovers from highly exposed ports to the rest of the country rather than
redistribution.50
The second test of extensive margin effects categorizes ports as “Entering” into
international trade if there is no exporting activity in the pre-crisis period and positive
exports in the post-crisis period, and “Exiting” if the reverse is true. I estimate a linear
probability model on a one-period cross-section of all ports where E po is an indicator for
either Entry or Exit and standard errors are clustered by the origin-country:
Pr(E po ) = a + bFail po + go + G0 X po + # po (1.9)
The full sample of 377 ports active in either period is 30 percent larger than the baseline
sample of intensive margin ports (those active in both periods). 52 of the new ports were
entries and 36 were exits, which implies a high degree of turnover in this window.51
Table 1.6, columns 4 and 6 present the within-country likelihood of Entry po and of Exit po ,
respectively. The point estimates are economically and statistically significant for Entry
and not significant for Exit. A port exposed to the average level of bank failures was 2.4
percent less likely to begin exporting at all.52
1.5.3 Limited within-country substitution
Having established that exposure to bank failures caused large intensive and extensive
margin declines in shipping at the port-level, I next address whether the contractions in
the local economy had aggregate implications at the country-level. To what extent could
50 These results are analogous to the findings in Huber (2018) that firms within a county that did not
directly experience a financing shock still performed worse post-crisis from declines in aggregate demand.
51 Ports likely remained active in domestic, coastal trade. However, the Lloyd’s List did not track nor
report on these types of ship movements.

52 Similarly, Berman and Héricourt (2010) find that access to finance influences the firm entry decision,
but that it has no effect on the exit decision.
43
Table 1.6: Extensive margin effect of exposure to bank failures
Port destinations Country destinations I(Port Entry) I(Port Exit)

(1) (2) (3) (4) (5) (6) (7)
Failpo ⇥ post -0.225** -0.295***
[0.112] [0.113]
Failo ⇥ post -0.484***
[0.163]
Failpo -0.161*** -0.193** 0.143 0.137
[0.0499] [0.0806] [0.123] [0.159]
Port controls ⇥ post Y Y
Port controls Y Y Y Y
Portp FE Y Y
Countryo ⇥ post FE Y
Country controls ⇥ post Y
Countryo FE Y Y Y
N 574 574 108 377 377 377 377
Ports 286 286 377 377 377 377
Clusters 54 54 54 55 55 55 55
Notes: Table 1.6 reports estimates of the effect of the exposure to bank failures on the extensive margin of
shipping activity. The dependent variable in columns 1 and 2 is the ln number of unique destinations
accessed by ports. The dependent variable in column 3 is the ln number of unique destinations accessed by
countries. The sample in columns 1 to 3 is restricted to ports that were active in both the pre-shock and the
post-shock periods. The dependent variable in columns 4 and 5, "I(Port Entry)" is a binary variable that
takes the value of 1 for a port that was not active in the pre-shock period and became active in the
post-shock period, and 0 otherwise. The dependent variable in columns 6 and 7, "(Port Exit)" is a binary
variable for a port that was active in the pre-shock period and became inactive in the post-shock period.
The sample in columns 4–7 includes all ports that were ever active in the year around the crisis.
All variables are defined the same way as in Table 1.7. Standard errors in brackets are clustered by country
of origin. *p < 0.1, **p < 0.05, ***p < 0.01
44
exporters ship from a neighboring port? I first estimate the relationship between country-
level exposure and shipping, and then I directly estimate the degree of substitution
between ports.
I aggregate shipping activity across ports within a country and estimate the country-
level analogue of the baseline DD in Equation 1.7. The dependent variable is ln(Sot )
where Sot is the total number of ships departing a country per period (Sot = Â p S pot ). Failo
is calculated according to Equation 1.3 from country-level shares of pre-crisis dependence
on individual banks. go controls for time-invariant country-level characteristics. b is
identified off across-country variation in the exposure to bank failures, so it is not possible
to control for origin-country time trends. However, I do control for pre-crisis country-level
characteristics that are correlated with the degree of bank failure. Table 1.7 presents
the baseline estimation with the full set of controls and directly compares the port and
country-level outcomes. The baseline coefficient in column 2 at the port-level is -0.558,
which is almost identical to the coefficient in column 4 at the country-level of -0.595.
These estimations reaffirm the patterns shown in Figure 1.3b. Table A4 reports robustness
to controlling for all the country-level characteristics.
Next, I directly estimate the degree of substitution among ports in a country by

asking whether more exposure to bank failures in the rest of the country benefits a
port, controlling for its own exposure. I construct a measure of the average exposure
to bank failures in the cities in the rest of the country, leaving out the port’s own city
of financing.53 Since it is a city-level measure, it is not collinear with the origin-country
trends.54 I include this measure as an additional control to Equation 1.7:
ln(S pot ) = bFail po ⇥ Postt + yFailother,o ⇥ Postt + a p + got + G0 X pot + # pot (1.10)
53 This measure is calculated by removing each city’s contribution from the country-level exposure
measure. A measure calculated by port would double-count cities that financed more than one port and
generate variation based on the number of ports rather than variation from the differences among cities.
54 I also estimate specifications without g where Fail
ot other,o ⇥ Postt proxies for origin-country trends.
45
Table 1.7: Immediate effect of bank failures on port- and country-level shipping
Port Country Port

(1) (2) (3) (4) (5) (6) (7)
Failpo ⇥ post -0.713*** -0.558*** -0.711*** -0.722*** -0.568***
[0.274] [0.189] [0.248] [0.153] [0.187]
Failo ⇥ post -0.505** -0.595**
[0.223] [0.251]
Failother p,o ⇥ post 0.0912 -0.421 -0.311
[0.0715] [0.426] [0.401]
ln(sugar) ⇥ post Y
non-Brit banks ⇥ post Y Y Y Y
Port controls ⇥ post Y Y Y
Portp FE Y Y Y Y Y
Countryo ⇥ post FE Y Y Y
Countryo FE Y Y
post FE Y Y
N 578 578 108 108 578 578 578
Ports 289 289 289 289 289
Clusters 54 54 54 54 54 54 54
Notes: Table 1.7 reports estimates from the difference-in-difference regressions from the two-period panel of
port-level shipping activity and country-level shipping activity in the year before and after the crisis. In
Columns 1–2 and 5–7, the dependent variable is the ln of the total number of ships departing each port in
each period; in Columns 3–4 it is the total number of ships departing each country in each period. Fail po is
the share of the port’s banks that failed during the crisis, Failo is the share of the country’s banks that
failed, and Failother,po is the country-level share of bank failures outside of port p. post is a dummy for the
post-crisis year. The port controls consist of an indicator for the port being a capital city within the country,
the average ln age of banks, the number of non-British banks, and the fraction of shipping to the UK. The
country controls consist of the ln of the value of sugar exports in 1865, and the number of non-British
banks. Countries that did not export sugar are given a given a ln value of 0. All controls are interacted
with the post dummy. The sample is restricted to ports ship in both the pre- and post-period. Standard
errors in brackets are clustered by country of origin. *p < 0.1, **p < 0.05, ***p < 0.01
46
y is the main coefficient of interest. It controls for a port’s own exposure to bank failures
and measures the semi-elasticity of its own exports to the rest-of-country exposure
to bank failures. y > 0 indicates that a higher degree of exposure in the rest of the
country benefits a port, and it implies that exporters from the rest of the country can find
alternative financing in the port. y > 0 would suggest that this channel of within-country
substitution could reduce the country-level losses. In Table 1.7 column 6, I report a
negative coefficient of 0.311. This estimate is not statistically significant, but it contributes
further evidence that exporters were not able to relocate within-country. It provides
suggestive evidence that city-level shocks had negative spillovers to the rest of the country.
1.6 Long-term impact on trade
The previous section showed that British bank failures negatively impacted exports
immediately after the crisis. In this section, I examine the long-run effects of the temporary
financing shock using the full panel of country-level values of exports from 1850-1914.
First, I use the across-country variation in exposure to document the persistent effects
then on the total values of exports and on bilateral trade relationships. Second, I explore
two channels that lengthened the recovery process.
1.6.1 Baseline results across countries
Total exports
First, I show the patterns of divergence in the raw data. In Figure 1.4a, I plot the annual
aggregate values of exports for countries binned into above and below-average exposure
to bank failure, where the average exposure is defined in the cross-section of countries,
and levels for each group are indexed to equal 1 in 1866.55 This figure shows that before
55 The patterns are almost identical using medians. Binning the countries into two groups is equivalent to
using a standard DD estimator with a treatment and control group.
47
1866, exports were expanding at the same rate between the two countries so there are
no differential pre-trends between the groups, but after 1866 there is an immediate
divergence in levels that does not recover.56 Figure 1.4b graphs the difference between
the two groups, which corresponds to the DD estimate with binary treatment.
The permanent divergence arises from a temporary jump in the annual exports growth
rates of unaffected countries in the four years after the crisis. In Figure A8, I plot the
annual growth rate of exports and show that they are very similar pre-crisis, diverge
after the crisis in 1867, and then converge again to the same pattern by 1880. In the
pre-crisis period, the average annual growth rates are 12 and 11 percent for the less
exposed (solid line) and more exposed groups (dashed line), respectively. This difference
is not significant; the p-value for difference in means is 0.77.57 In 1867 the less exposed
group (solid line) grew 31 percent while the more exposed group (dashed line) grew 6
percent, and in 1868 the growth rates were 21 and 12 percent respectively. The cumulative
difference in the annual growth rates between the two groups after the first two years
is 33.6 percent. This initial difference in export growth rates is the main driver of the
average annual difference in growth rates of 1.8 percent per year between groups from
1867–1914.58
Next, I benchmark these findings against estimates of the elasticity of trade with
respect to geographic distance. Using my dataset, I estimate a trade elasticity of -1.1 to
geodesic distance.59 Relative to this elasticity, increasing an exporter’s exposure to bank
failures from below to above average is equivalent to increasing its geographic distance
56 The country-level divergence shown here is another piece of evidence that there was little within-country
reallocation of exporting activity.
57 In the immediate pre-crisis period from 1860–1865, the average annual growth rates were 6.4 and 6.0
percent, respectively, and the p-value for the difference in means is 0.92.
58 The average annual growth rates from 1867–1914 are 4.5 and 2.7 percent for the less exposed and more
exposed groups of countries, respectively. This is calculated using the 1914 values of exports, which were
8.47 and 3.59 times the values in 1866 for the two groups, respectively.
59 In other words, a 1 percent increase in physical distance between two countries reduces the trade flows
between them by 1.1 percent. This elasticity is, coincidentally, exactly the average elasticity found in the
literature based on the survey of structural gravity by Head and Mayer (2014). It is slightly larger than
the average estimate of -0.93 found in all gravity papers. Table A6 reports the estimates and robustness to
controlling for gravity measurements of bilateral resistance.
48
Figure 1.4: Aggregate exports, grouping countries by above and below average exposure to bank failures
(a) Exports by group
10
Exports (1866=1) 8
0
1845 1855 1865 1875 1885 1895 1905 1915
Year
Below avg failure Above avg failure
(b) Difference between groups
0
Above avg exposure - Below avg exposure
-2
-4
-6
1845 1855 1865 1875 1885 1895 1905 1915
Year
Notes: Figure 1.4a plots the raw data for the total value of exports by groups of countries from 1850–1914.
Countries are binned into two categories: “Below avg failure” refers to countries that experienced below
average exposure to bank failures in London, where the average rate was calculated in the cross-section of
exporting countries in 1866. “Above avg failure” refers to countries that experienced above average
exposure to bank failures. Exports values are normalized to equal 1 in 1866. Figure 1.4b plots the
difference between the values for the two groups. The vertical line marks 1866. Figure A7 plots the
coefficients and standard errors from the equivalent regression.
49
to its trading partners by 30.6 percent after the first two years. As a concrete example of
the magnitudes, if Spain only exported to the United States, then above average exposure
to the shock is equivalent to moving Spain over 1,400 miles to modern-day Turkey.60
The impact on exports is much larger than the impact on GDP, although there also
appears to be a permanent effect on GDP levels. In Figure A9, I plot aggregate GDP for
the same two groups of countries, binned by above and below average exposure to bank
failures. The difference in the average annual growth rates in output is only 0.6 percent,
which is one third of the difference for exports. As in the Great Trade Collapse of 2008,
the difference in exports is much larger than the difference in GDP, so the trade-specific
losses cannot be driven by productivity declines that affect output as well.
Bilateral exports
I formally estimate the effect of exposure to bank failures on bilateral exports with Equa-
tion 1.8, which allows for demand shocks in the form of destination-country-year fixed
effects. I allow b t to vary annually and at five-year intervals ([1850, 1855], ..., [1911, 1914]).
b t should be interpreted as the semi-elasticity of the response to exposure to bank failures
in the exporting country by a given importer in a given year. For example, how much
less is France predicted to import from Chile (20 percent exposure) than Brazil (2 percent
exposure) in the year 1900?
Figures 1.5a and 1.5b plot the estimated b t coefficients annually and at five-year inter-
vals, where b 1866 and b 1861 5 are the omitted categories in each specification, respectively.
b t reflects the relative exports in the cross-section with a continuous measure of exposure
and therefore does not necessarily imply a drop in the aggregate levels of world trade.
The estimated coefficients support the patterns in the raw data that exposure to the crisis
had no effect on exports pre-crisis, but that it immediately lowered trade flows between
countries afterward. I report the point estimates in Table A7 (Column 2).
60 Thesedistances are the shortest route between the geodesic centers of each country. The distance
between the US and Spain is 4,715 miles, and between the US and Turkey is 6,327 miles.
50
Figure 1.5: Financing shock has long-term effects on exports

(a) b t estimated annually
2
βt: treatment effect of exposure
-2
-4
1850 1860 1870 1880 1890 1900 1910 1914
Years
(b) b t estimated every 5 years
1
-1
-2
-3
5 0 5 0 5 0 5 0 5 0 5 0 11-14
-185 56-6 61-6 66-7 71-7 76-8 81-8 86-9 91-9 96-0 1901-0 06-1
1850
Years
Notes: Figure 1.5 plots the b t point estimates and 95 percent confidence intervals for the specification given
in equation 1.8 estimated on the country-level panel of trade. The dependent variable is the ln value of
exports. The specification includes origin country o FE, destination country-year dt FE, and time-varying
controls for the bilateral distance between countries. b t is the treatment coefficient on the effect of exposure
to failed banks on exports in each group of years. Standard errors are clustered by the origin country. See
Table A7 column 1 for the point estimates. N = 67,378.
51
The persistence is striking: destination countries imported less from exporters that
had been exposed to bank failures for almost 40 years. The average estimated annual
coefficient from 1867–1900 is -1.71 log points. b 1901 05 is the first period when the effect
is not statistically different from zero. However, the average magnitude of the coefficients
after 1900 is -1.11, which is still 65 percent of the average effect until 1900. The average
estimated coefficient from 1867–1914 is -1.53 log points, and given the average exposure
of 11 percent, implies that the (partial equilibrium) reduction in world exports during
this period was 17 percent per year.
I expand the bilateral estimation to encompass all years from 1850–2014 and plot
the estimated coefficients and 95 percent confidence intervals in Figure A10 (coefficients
reported in Table A8 Column 2), marking the years corresponding mostly closely to
the two world wars. The full time horizon shows that there is a very slow pattern
of convergence, with coefficients mostly not statistically different from zero after 1930.
However, the estimated coefficient in the final period, b 2011 14 is -0.81, which is 53 percent
of the estimated effect from 1867–1914.61
The burden of the losses falls on new trade relationships that had not existed before
1866. In Figure A13, I categorize bilateral relationships by whether they are new or pre-
existing, and I show that the same exporters had larger losses in their new relationships.
This result is consistent with the institutional context in which banks provided the
financing that overcame initial contracting frictions between importers and exporters.
It also suggests the persistent effects can be driven by the early loss in market share,
and that country characteristics that would protect them from those losses would also
generate faster recovery.
61 TableA8 Column 3 shows that among the original group of countries that were active in international
trade in 1866, the magnitudes of the effects are as large and statistically significant in 2014 as in the
pre-WWI period.
52
Robustness
I test the robustness of the long-term results by controlling for observable characteristics
that could be confounding factors, and by implementing the Fisher exact test.
In Table A7 Columns 3–8, I show robustness to a variety of origin-country controls,
including the pre-crisis characteristics that are correlated with bank failures. In Table
A9 I report the estimates after including standard gravity covariates, such as shared
language, shared land border, and being in the same European empire. Additional
robustness includes controlling for pre-crisis and contemporary military conflicts (Table
A10); exchange rate regimes pre-crisis (Table A10); industry composition of exports pre-
crisis (Table A11); financial crises like sovereign debt, domestic debt, stock market crashes
both contemporaneous and in 1865 (Table A13 and A14); and ability to issue long-term
debt or equity in London (Table A15).62 The static and the time-varying versions of all of
these controls do not affect the statistical significance or the qualitative patterns of the
results.
I also test the robustness of the long-term results by implementing the Fisher exact test
for randomization inference. This test is conducted by reassigning treatment randomly
without replacement to compare the estimated treatment effect against hundreds or
thousands of placebos. This test is one way to check for the possibility that at longer
time horizons, countries diverge for other reasons, and the bank exposures are correlated
with those long-term changes. Assigning the treatment randomly will show whether
the long-term negative effects could arise naturally from the data for reasons unrelated
to the banking shock. If that is the case, the distribution of estimated coefficients will
become more negative left with each subsequent group of years. If there is no such drift,
the distribution should remain around zero, as is the case in randomization tests in the
cross-section.
62 It
is only necessary to control for characteristics in the origin-country or between country-pairs because
the baseline specification includes destination-country year fixed effects, which will absorb conflicts
occurring in the destination country.
53
In this test, I redistribute the shocks randomly and simulate the data 1,000 times,
then estimate the long-term effects in Equation 1.8 using the simulated data. I plot the
distribution of the coefficients for each group of five years in Figure A14. These plots show
that the coefficients are centered around zero in all periods. The lack of drift suggests
that the long-term effects are not likely to have been generated by unobserved processes
of divergence.
Banking sector recovery
A natural explanation for the persistent effects is that the banking sector does not recover.
Given British banking dominance, the shock in London could have caused a permanent
retrenchment in multinational banking, especially in the locations most affected by bank
failures. I test this hypothesis explicitly using the city-level panel of banks. I find that
multinational banking did not retrench: Figure A11 shows that the global distribution
of banks became consistently more widespread and denser with time. I plot the total
number of banks and the composition of banks by nationality at the city-level by above
and below average exposure to banks that failed in figure A12. Figure A12a shows that
cities that were more exposed to bank failures had access to the same number of banks as
cities that were less exposed. This figure shows that the persistent effects across countries
could not be explained by the size of the banking sector, measured by the number of
banks.63
While there is no difference in the total number of banks, there is a change in the
composition of nationalities among banks. Figure A12b shows that British banks did not
tend to return to the locations that had experienced a higher degree of failures, but that
domestic and other European banks filled the gap, likely responding to the credit supply
gap left by British banks (Figures A12c and A12d). These patterns are consistent with the
63 A full time-series for the balance-sheet characteristics of all the banks is not available. The balance-
sheets for a subset of banks are available in 1901, which I use to verify that banks are of similar average
size across nationalities.
54
historical consensus that after 1870, France and Germany actively sought to expand their
financial presence around the world to compete with Britain (Einzig, 1931; Kisling, 2017).
However, I formally control for the number of banks of different nationalities and show
that these do not alter the persistent effects in the baseline results (Table A16).
1.6.2 Channels for persistence
Having established that exposure to bank failure affects economic activity in the long-run,
I explore two channels for the persistent effects: exporters’ lack of access to alternative
forms of financing and importers’ ability to substitute to less credit-constrained exporters.
These two channels are trade-specific mechanisms that would address the relative decline
in exports relative to output.
Access to alternative financing
Exporters who had more than one banking relationship would have been able to source
some credit from these other relationships. The presence of non-British banks could have
provided an alternative source of financing that may mitigate the main effects of bank
failures. In appendix A.4, I present additional evidence that exports were less affected in
trade relationships that can substitute away from British financing.
Immediate effects within countries

I use the port-level panel to test this hypothesis in the short-term using within-country
variation. I do not observe non-British financing relationships directly so I proxy for
them using the number of non-British banks pre-crisis. I re-estimate Equation 1.7 with an
interaction term between exposure to failure and the number of non-British banks:
ln(S pot ) = bFail po ⇥ Postt + fFail po ⇥ non-Brit po ⇥ Postt + a p + got + G0 X pot + # pot (1.11)
f is the main coefficient of interest: f > 0 means that conditional on exposure to bank
55
failures, exports were higher in ports that had access to non-British banks. Table 1.8
(Column 2) confirms that having access to more non-British banks pre-crisis mitigated
the main losses. At the port-level, there is no correlation between the number of non-
British banks and the likelihood of bank failure, so this result is not driven by any trends
correlated to non-British banks. The magnitude of f (non-Brit banks ⇥ Fail po ⇥ post) is
34 percent of the baseline effect. The estimated coefficient is statistically significant at
the 1 percent level, but the economic magnitude depends on assumptions about the size
and effectiveness of non-British banks relative to British banks in providing trade credit.
Assuming the same size and effectiveness, the average port had access to 0.6 non-British
banks, which means that access to other bank-intermediated finance mitigated the main
effect of exposure to bank failures by 20 percent.
Table 1.8: Port access to alternative sources of financing
(1) (2)
Failpo ⇥ post -0.936*** -0.805***
[0.227] [0.240]
non-Brit banks ⇥ Failpo ⇥ post 0.290*** 0.270**
[0.111] [0.106]
non-Brit banks ⇥ post Y Y
Port controls ⇥ post Y
Countryo ⇥ post FE Y Y
Portp FE Y Y
N 578 578
Ports 289 289
Clusters 54 54
Notes: Table 1.8 reports estimates of the effect of access to alternative forms of financing on shipping
activity. The dependent variable is the ln of the number of ships sailed. non-Brit banks is the number of
non-British banks in the port’s city of financing in the pre-crisis year. All other variables are defined the
same way as in Table 1.7. Standard errors in brackets are clustered by the origin-country. *p < 0.1, **p <
0.05, ***p < 0.01
Long-term effects across countries

I estimate the long-term effects of gaining access to alternative banking networks by using
56
the nationalities and identities of the multinational banks within each city in the five year
windows from 1850-1914. French and German banks are the most important alternatives
because they accessed the second and third largest money markets in the world after
London, and were created to compete with British banks (Einzig, 1931; Kisling, 2017).
I construct a binary variable called “European bank” (I (EBo )) that takes the value
of 1 when the exporting country has access to either a French or German bank, and 0
otherwise. This variable proxies for access to the most likely alternative to the London
money market. I estimate the following:
ln(EXodt ) = qt Failo ⇥ I (EBod ) + b t Failo + lt I (EBod )

(1.12)
0
+ Y Xod + go + gdt + # odt
lt absorbs the time-varying effect of access to common banks across all countries. Xod are
standard gravity variables of bilateral resistance.64 Figure 1.6 plots b t in orange and qt in
blue. Interacting I (EBo ) with the exposure to failure each year estimates the additional
effect of access to alternative financing for exposed places. The full effect for exposed
places is qt + b t , which is close to 0 for most years, indicating that countries without
access to other financing networks are the ones driving the main losses seen in Figure 1.5.
Exports substitutability
In this section, I depart from financial frictions and discuss frictions arising from competi-
tion in exports markets. A trade cost shock between parties can lead importers to source
from new relationships or to increase the amount they buy from pre-existing relationships.
In the 19th century, most countries exported commodities that were produced by multiple
other countries, leading to a high degree of substitutability across countries. As an
example, a country importing sugar could choose among a number of producers in the
Caribbean and South America. Countries exporting the same goods can therefore be
64 The results are robust to not including them and to allowing them to vary over time.
57
Figure 1.6: Recovery is better with access to other banks
ln(EXodt ) = qt Failo ⇥ I (EBod ) + b t Failo + lt I (EBod ) + Y0 Xod + go + gdt + # odt

βt: treatment effect & θt: interaction with European bank
-2
-4
-6
5 0 5 0 5 0 5 0 5 0 5 0 11-14
-185 56-6 61-6 66-7 71-7 76-8 81-8 86-9 91-9 96-0 1901-0 06-1
1850
Years
Failure x European bank Failure
Notes: Figure 1.6 plots the b t and qt point estimates and 95 percent confidence intervals for the specification
given in equation 1.12 estimated on the country-level panel of trade. The dependent variable is the ln value
of exports. The specification includes origin country o FE, destination country-year dt FE, time-varying
controls for the bilateral distance between countries, and time-varying indicators for common land border,
common European colony, and common language. “Failure ⇥ European banks” is the interaction effect of
exposure to failed banks on exports in countries with access to other European banks. “Failure” is the
treatment effect of exposure to bank failures for all countries. Standard errors are clustered by the origin
country. N = 67,378.
58
modeled as homogeneous firms with different variable trading costs. A large shock to
the cost of exporting from one country can lead competing exporters to enter into that
country’s markets.
First, I use the industry composition of a country’s exports pre-crisis, categorized
by two-digit SITC codes, to test for importer substitution among similar countries. The
global value of exports by SITC is shown in Figure A1. I calculate the top SITC group by
geographic region and include these as time-varying controls. This estimation is restricted
to the 44 countries with the exports composition, so the results are noisy, but they indicate
no recovery.
Next, I proxy for similarity in exports products using each country’s geographic region
to include countries where product-level exports data are not available. I validate that
geographic region is a reasonable proxy for the goods exported by evaluating the proxy
on the subset of 44 countries with observable industry composition in 1865. For each
region, I identify the top three exports categories by SITC codes and calculate the fraction
of the total value of exports from the region that fall into those categories.65 This fraction
is equivalent to an exports-weighted average of the cross-country exports concentration
within the top three categories. Figure A15 shows that this fraction is above 0.5 for all
regions and averages 0.73 across regions, indicating that exports are very similar within
region.
I compare the countries within regions to each other by including origin-country
region-year fixed effects in the baseline specification in Equation 1.8. The additional
controls restrict the variation such that b t is estimated off comparisons of countries in the
same geographic area exporting to the same destination in the same year. Figure 1.7 (Table
A7 Column 8) shows that there is no recovery in this setting. The qualitative interpretation
is that within regions, countries that are more exposed to bank failures experience exports
losses for longer than the other countries in the group. I also re-estimate the baseline with
65 Eachregion has at least two countries, and the primary exports for all countries outside of Northwest
Europe are raw commodity goods.
59
Figure 1.7: Recovery is worse within groups of countries with similar exports
ln(EXodt ) = b t Failo + G0 Xot + go + gdt + yRegionot + qt ln(dist)od + # odt
1
βt: treatment effect of bank failure exposure
-1
-2
-3
5 0 5 0 5 0 5 0 5 0 5 0 11-14
-185 56-6 61-6 66-7 71-7 76-8 81-8 86-9 91-9 96-0 1901-0 06-1
1850
Years
Notes: Figure 1.7 plots the point estimates and 95 percent confidence intervals for the specification given
above estimated on the country-level panel of trade. The dependent variable is the ln value of exports. The
specification includes origin-country region-year FE, origin country o FE, destination country-year dt FE,
and time-varying controls for the bilateral distance between countries. b t is the treatment coefficient on the
effect of exposure to failed banks on exports in each group of years. Standard errors are clustered by the
origin country. See Table A7 column 8 for the point estimates. N = 67,378.
region-year fixed effects using the subsample of countries that have SITC information
and verify that the patterns are similar. The coefficients are plotted in Figure A18, and
the point estimates are reported in Table A7.
Second, I test for positive spillovers within region by estimating the effect of other
countries’ average exposure on a given country’s exports, controlling for that country’s
own exposure. The prediction is that there should be positive spillovers because a trade
cost shock to certain countries will benefit their competitors with similar exports. I find
evidence of positive spillovers (Figure A17), but the estimates are noisy.
The sustained persistence of the effects within regions are not driven by the smaller
sample comparisons. In a robustness check, I conduct a Fisher exact test for the country
60
groups by simulating 1,000 random group assignments and re-estimating the coefficients.
I plot the distribution of the five-year coefficients in Figure A16. This figure shows that
the true estimates are very similar to the simulated estimates for the years until 1900.
At that point, the true coefficients are larger in magnitude than the average simulated
coefficient. These results suggest that substitution in real goods markets, where importers
sourced from less exposed countries that could provide similar goods, can explain the
persistent effects
1.7 Conclusion
Standard macro-finance and trade models imply that financial crises only affect the real
economy as long as the financial sector has not recovered, yet crises lasting just a few
years have been correlated with declines in GDP and trade lasting at least a decade. This
paper uses a salient historical setting and novel archival data to provide new causal
evidence on the real economic effects of bank failures in the long-term. The most severe
banking crisis in British history serves as a laboratory where London’s role as the global
financial center meant that bank failures in London were exported abroad to cities and
countries around the world. Exposure to bank failures caused large immediate declines
in exporting activity on both the intensive and extensive margins within and across
countries, and that the country-level losses persisted for almost four decades.
The main contribution is to document that even a short-lived financing shock can
lead to persistent divergence in the geographic distribution of economic activity. The
persistent effects are driven by countries without access to alternative bank networks
and by those in more competitive exports markets. First, having access to non-British
banks mitigates one third of the losses in the short-term and almost all of them in the
long-term. Second, the countries whose competitor in major exports markets were highly
exposed to the bank failures benefited. Within groups of countries exporting similar
61
goods, more exposed exporters had no recovery by 1914. This hysteresis empirically
documents the theoretical argument that one-time trade cost shocks can permanently
affect the distribution of trade activity (Baldwin and Krugman, 1989).
The results in this paper contributes to our understanding of the real costs of financial
crises, especially in the long-run. The slow post-crisis recovery among advanced countries
in recent decades suggests that the historical record is more relevant than ever. It also
provides further evidence that international trade is a sector particularly sensitive to the
costs of external finance, but it highlights how short-term changes to trade costs affect
long-term trade relationships. While this paper focuses on the impact of losing banks that
intermediated trade, it also showed that having access to other forms of finance mitigated
the long-term losses. Gaining an understanding of how access to finance expanded trade
networks in both the current and First Ages of Globalization would be a fruitful avenue
for future research.
62
2 | National Banks and the
Liabilities Channel of
Local Economic Activity1
2.1 Introduction
There is a large literature on the importance of financial intermediaries in economic

growth both across countries (e.g. Rajan and Zingales (1998b)) and within the United
States (e.g. Jayaratne and Strahan (1996)). Causally establishing the specific role that banks
play in promoting growth is challenging because financial institutions are not allocated
randomly and the local economic conditions will be correlated with characteristics of the
bank itself. In addition, most of the empirical evidence in this literature has focused on the
assets side of the balance sheet through the bank lending channel, yet many innovations
to the banking sector have been on the liabilities side with, for instance, the introduction
of deposit insurance. In the modern economy, there is no variation in the quality of the
two most common forms of bank debt—bank notes and demand deposits.2 While several
theoretical papers have shown that banks can issue liabilities that are sufficiently riskless
that they can act as a transactions medium (Gorton and Pennacchi, 1990; Stein, 2012),
1 Co-authoredwith He Yang
2 The
former is printed by the central bank with some exceptions such as Scotland and Northern Ireland
whose currency is supplied by several private institutions, and the latter is government-insured.
63
there is little empirical evidence demonstrating their effects on the real economy.
In this paper, we explore the effects of secure bank liabilities on the real economy by
studying the historical period after the National Banking Act of 1864. The Act created a
new class of federally regulated “national banks." With the aim of promoting banking
sector stability, regulatory capital requirements were set for national banks based on
town population, creating a plausibly exogenous entry cost for new banks in towns near
the population cut-off. In addition, it mandated that all national bank notes could be
redeemed for their nominal value. These two characteristics of the historical setting
therefore provide a natural experiment for local economies gaining access to stable bank
liabilities that could better serve as a medium of exchange to facilitate trade and growth.
More specifically, the first key regulatory difference between national banks and
pre-existing “state banks” was the capital requirements based on the population of the
town in which the bank was incorporated.3 In towns with fewer than 6,000 people
the capital requirement was $50,000, and in towns with 6,000-50,000 people the capital
requirement doubled to $100,000.4 The discontinuous jump means that towns just below
the population cutoff at 6,000 faced significantly lower entry costs for establishing a
national bank than towns just above the cutoff. These requirements were directly relevant
for local national bank entry as the banks were not allowed to branch. In addition, national
banks were more circumscribed in their lending practices, which were, in conjunction
with the higher equity capital requirements, designed to make them more stable.
The second key regulatory difference was that national banks were required to circulate
a standard currency, which grew out of the concerns that uncertainty over the value of
bank notes depressed economic activity (e.g. Cagan (1963); Knox (1900)). Prior to the
establishment of national banks, state banks printed their own notes that only circulated
3 Statebanks were so called because they were chartered under state legislation, which in general were
more lax than the newly introduced federal legislation.
4 The capital requirement was based on the population of incorporated localities, which could be cities,
towns, boroughs, or villages in some states. For simplicity we will refer to all localities as “towns” going
forward. The capital requirement was $250,000 in towns with population above 50,000.
64
locally and traded at discounts relative to one another (Ales et al., 2008; Gorton, 1999).
The varying discounts introduced spatial and temporal price uncertainty that raised
transactions costs. By gaining access to a uniform currency, towns increased their “market
access” by reducing trading cost with distant places.5 We interpret the introduction of
national banks as a shock to a town’s market access that should positively affect its trade
activity.
We use the different regulatory capital requirements as an exogenous source of
variation in gaining a national bank by instrumenting for bank entry with town population
being below the 6,000 threshold, according to the most recently published census. We
focus on the towns that gained a national bank for the first time in order to capture the
impact of directly accessing the more stable national bank notes. Our analysis begins
with the 1870 census rather than 1860 one to allow for the disruptions caused by the Civil
War to equilibrate.
Our sample consists of towns that had fewer than 6,000 people in 1870 and no national
bank as of 1875, and had between 4,000 and 8,000 people in 1880.6 Choosing a small
population bandwidth of 4,000–8,000 allows us to limit our sample to towns that are
likely to be similar in both observable and unobservable characteristics. Among these
similar towns, some crossed the 6,000 population cutoff in the 1880 census, which doubled
the entry cost for a national bank. The identifying assumption is that there was not
a concurrent shock to lower-populated places that would cause them to grow faster
after 1880. We control for observable differences in their growth trajectory as proxied
by population growth, a towns financial development in the pre-period as proxied by
the number of state banks in the town, as well as the area’s transportation condition as
proxied by the number of railroads in the county. The predicted first stage is that towns
5 The term market access is a reduced form expression for the costs of trading with all other places. It
arises from general equilibrium trade theory, and the market access for a given location is a function of the
sizes and trading costs of transacting with all other locations.
6 The analysis is not sensitive to choosing 1875 as the first year we calculate whether a town has a
national bank. We show robustness to other years in the appendix.
65
that did not cross the population threshold should be more likely to establish a national
bank.
We show that our instrument strongly predicts the likelihood of a town having a
national bank in the mid-1880s: having fewer than 6,000 people in 1880 is associated with
30% higher probability of gaining a national bank as of 1885.7 Controlling for a town’s
growth trajectory in the pre-period greatly alleviates endogeneity concerns with the
relationship between population levels in 1880 and future growth in these places. We also
show that pre-period observable characteristics were not significantly different between
the towns with a population above 6,000 versus those had below 6,000 population in 1880,
both conditionally and unconditionally. The balance on observables further suggests that
selection on unobservable characteristics is less of a concern.
Having established that lower regulatory capital requirements increased bank entry,
we first show that national bank entry significantly shifted the composition of goods
produced toward traded goods while not affecting total agricultural production. Output
of traded crops (those listed on the Chicago Board of Trade), such as wheat and oat,
crowded out non-traded crops, such as rye and barley. The pure compositional change
avoids confounding effects from other bank activities that would also affect the levels
of production. It suggests that the shift towards traded products was due to the access
to a stable currency that reduced transactions frictions. We also find that national bank
entry did not impact agricultural capital that was financed by short-term credit (proxied
by fertilizers expenditure) or long-term credit (proxied by the value of land and fixtures
on farms). National bank lending to the agricultural sector was overall limited, which is
consistent with the evidence that there were restrictions against lending to agriculture
(Knox, 1900).8
7 This finding are consistent with Fulford (2015), Gou (2016) and Carlson et al. (2018) which also use
the population cut-offs governing national banks’ regulatory capital requirement in the late 19th and early
20th centuries. This result follows an older literature by Sylla (1969) and Sylla (1982) that the high capital
requirements imposed by the National Banking Act hindered financial development.
8 The call reports do not provide information on the nature of the loans the banks held, so it is not
possible to test this directly.
66
We further demonstrate the positive impact of national banks on trade using direct
measures of local trade activity as proxied by employment in trading sectors. In particular,
we use contemporary business directories with town-level coverage and the full count
Censuses of Population to show that there was a relative increase in the number of
commission merchants, buyers, and shippers in places that gained national banks. We
also show that placebo professions that might capture general growth, such as architects
and doctors, did not differentially benefit from gaining a national bank.
Next, we study the impacts of national banks on the manufacturing sector, as manufac-
turing products are tradable and therefore the sector likely benefited from the increased
trade activity following national bank entry. We use the decennial Census of Manufac-
tures to show that places that gained a national bank experienced significantly greater
growth in total production between 1880 and 1890. Accessing national banks caused
counties to increase their manufacturing output by $310 per capita from 1880 to 1890
(about $8,620 in 2018 dollars) or about 50% relative to 1880 levels. These results are not
driven by the number of state banks before 1880 and the number of railroads in 1880.
The growth in manufacturing output appears to be driven by growth in inputs and
employment whereas manufacturing capital was not significantly affected. These results
suggest that national banks were unlikely to have facilitated growth through extending
long-term loans for capital acquisition (consistent with Fulford (2015)). Instead, outputs
growth was likely driven by short-term working capital and liquidity provision operating
through the assets side of the balance sheet, as well as transactions costs reduction in
inputs and outputs trading operating through the liabilities side.
Since the period we study was a part of the Second Industrial Revolution, when orga-
nized industrial R&D within firms emerged (Bruland and Mowery, 2006), we postulate
that the significant manufacturing production could also be attributed to innovation.
We use the number of patents at the county level as a proxy for manufacturing sector
67
R&D output.9 We find that places that gained national banks had higher increases in
innovation output, measured by number of patents granted to local inventors. One
possible explanation for this result is that the increased trade activity in places gained
a national bank led the manufacturers to greater market access and exposed them to
more varieties of products, which could therefore incentive innovation for more product
differentiation.
Lastly, we examine whether the positive effect of local national bank entry on manufac-
turing sector growth persisted over time, and find that the elevated levels of production
persisted until the 1900s.10 The initial growth in conjunction with the growth in trade
and innovation may contribute to the persistence. In particular, the short-term changes
in trade and innovation activity likely gave places with national banks a comparative
advantage in the manufacturing sector over a longer period.
Our results indicate that the more stable national banks played an important role in
local economic growth, and that the liability channel has first-order impact on production
decisions and economic development. However, given that the entry of a national bank
brought multiple changes in banking services, it is not possible to fully disentangle the
relative importance of asset- and liability-side channels. For example, without detailed
information on each national bank’s loan portfolio, we cannot rule out the possibility that
increased trade activity, proxied by changes in trade-related professions, may also be a
result from trade finance provided by national banks.
Our identification strategy follows a number of papers studying bank behavior in
the postbellum United States, and in particular those that use the regulatory capital
requirements based on population as a source of exogenous variation. For example,
Gou (2016) uses the introduction of a new population cutoff in the early 20th century to
9 We provide evidence that at the county level, the number of patents obtained by local residents was
strongly correlated with manufacturing production between 1850 and 1890, and therefore is a reasonable
proxy for manufacturing sector R&D in this period.
10 The county-level data on manufacturing do not exist for 1910, and the levels are no longer significantly
different by 1920. Given the large amount of changes that likely occurred in WWI, we stop our analysis on
the long-term effect at 1900.
68
study the effect of capital requirements on bank stability. Similarly to Fulford (2015) and
Carlson et al. (2018), we focus on the 6,000 cutoff in the 1880 census, and as in Carlson
et al. (2018) we also control for town population growth following the previous census
as a proxy for a town’s overall growth trajectory. Unlike Fulford (2015), we use a more
geographically disaggregated unit than the county, and in contrast to Carlson et al. (2018),
we focus on towns that gained a national bank for the first time rather than incumbent
entry.11
To the best of our knowledge, this paper is the first to provide empirical evidence that
secure and stable bank liabilities can positively affect real economic activity by acting
as a form of privately issued money. While the large literature on the importance of
bank debt liquidity has mostly focused on their fire-sale externalities and their role in
raising financial fragility (Admati and Hellwig, 2014; Diamond and Dybvig, 1983; Stein,
2012), this paper studies a context in which they are truly safe assets (backed by federal
bonds) and therefore create a Gorton and Pennacchi (1990)-type transactions medium.12
In addition, our focus is on the real economic effects rather than decomposing and pricing
the safety and liquidity premia for these assets (Krishnamurthy and Vissing-Jorgensen,
2012; Nagel, 2016). In that sense, we provide evidence for a sub-national monetary
channel in the tradition of Friedman and Schwartz (1965); Romer and Romer (1989).
Furthermore, this paper adds to the literature of the historical determinants of eco-
nomic growth in the United States. This literature has focused on various factors, such as
technological innovations that reduced transportation costs (e.g. Donaldson and Horn-
beck (2016)), information acquisition costs (e.g. Feigenbaum and Rotemberg (2014)), or
11 The concurrent work by Carlson et al. (2018) focuses on the role of bank competition and thereby
requires a different sample of towns. Their analysis primarily focuses on the asset side of the balance sheet
and shows implications for bank lending behavior, leverage, and survival during the 1893 Panic.
12 Quadrini (2017) develops and calibrates a multi-sector model to show how the financial intermediation
sector can impact productivity through the bank liability channel. In his paper, bank liabilities provide
insurance for economic agents who become less constrained in their investments, which is different from
our monetary channel. Kashyap et al. (2002) study the synergy between lending and deposit-taking in
liquidity provision, and Pennacchi (2006) shows that the federal safety net provided by deposit insurance
promotes banks’ ability with liquidity provision. Our paper does not study the implication of stable bank
liabilities on lending.
69
changes in demographics (e.g. Sequeira et al. (2018)). This paper is complementary, and
contributes to the better understanding of how the banking sector has impacted economic
activities in the late 19th century (Carlson et al., 2018; Fulford, 2015; Landon-Lane and
Rockoff, 2007; Rousseau and Sylla, 2005; Rousseau and Wachtel, 1998; Weiss, 2018).13 In
particular, our results are consistent with Sylla (1982), which argues that the National
Banking Act’s high regulatory capital requirements held back economic development by
failing to expand the bank note supply sufficiently.
The rest of the paper is organized as follows. Section 3.2 discusses the historical
context around the time period studied, and provides motivating facts for the assets
and liabilities channels. Section 2.3 explains the data collection, sample construction,
and the empirical strategy. Section 3.4 presents the empirical results on the effect of
national bank entry on real economic outcomes, such as production in the agricultural and
manufacturing sectors, local trade activity and innovation output. Section 2.5 concludes.
2.2 Historical context
In this section, we provide an introduction of the historical background before and after
the National Banking Act of 1864.
2.2.1 The Free Banking Era
Between the expiration of the charter of the Second National Bank in 1836 and the
establishment of the Federal Reserve system in 1913, there was no unified banking system
in the United States.14 The National Banking Act of 1864 marked an intermediate step
when federally-regulated banks operated alongside state-regulated banks. The period
13 Both Fulford (2015) and Carlson et al. (2018) attribute the impact of national banks to the credit channel
on the asset side of the balance sheet. While Fulford (2015) studies rural counties that gained national
banks for the first time, Carlson et al. (2018) studies the role of competitive entry on incumbent banks’
lending behaviors.
14 See Appendix B.4 for more historical background on the First and Second National Bank before the
Free Banking Era.
70
before the National Banking Act was the Free Banking Era. During the Free Banking
Era, bank were chartered by their states and were subject to different regulations. State
banks were prevented from branching, and interstate banking was forbidden. Regulatory
oversight was generally weak, and bank runs and failures were frequent (Grada and
White, 2003). There was no formal system of interbank lending nor a lender of last resort,
and the banking sector experienced regular periods of booms and busts.15 As a result,
the antebellum banking system was fragmented, loosely regulated, and local economies
were exposed to the conditions of their local banks.16
A well-known feature of the Free Banking Era was that banks issued their own bank
notes which were only redeemable at face value in specie at the originating bank’s office.
In 1860 on the eve of the Civil War, there were almost 1,600 state banks, each issuing
its own bank notes. In large cities, the notes from hundreds of banks circulated at any
given time. The fact that there were so many bank notes in different designs means that
they were hard to verify and subject to counterfeit. Publications known as “Bank Note
Reporters and Counterfeit Detectors" were crucial for determining the legitimacy of a
currency. Figure B1(A) displays an example of a private bank note from Massachusetts
with face value $20, where the name and location of the issuing bank is prominently
displayed. Figure B1(B) shows the written description for the same bank’s notes in a
printed “counterfeit detector," where the $20 bill is described in the bottom left corner.
The lack of regulatory oversight from state legislations meant that banks often issued
notes beyond their redemption capabilities, causing uncertainty in the value of their
bank notes.17 Due to the asymmetric information in bank notes value and the physical
redemption costs, the bank notes did not generally trade at par with each other, creating
15 There were two exceptions: the Suffolk Banking System that served New England banks between 1827
and 1858, and a nonprofit collective founded in 1853 (Weber, 2012).
16 See Rockoff (1991) for a comprehensive review of the key characteristics of the Free Banking Era.
17 Milton Friedman referred to the phenomenon of banks over-inflating their currency to the point of
not being able to meet redemption as ‘wildcat banking,’ a term that is now frequently applied to the
Antebellum period in American banking. Gorton (1996) shows that discounts on bank debt can also arise
from a lack of credit history.
71
numerous and constantly changing exchange rates among the currencies (Ales et al., 2008;
Gorton, 1999). The large volatility in state bank notes value over time also indicates
significant time-varying idiosyncratic bank risks. For example, Figure 2.1 plots average
discounts of state bank notes in several states relative to banks in Philadelphia.18 Despite
the relatively low cost of acquiring information on the state banks in nearby regions, the
discounts and premiums were as high as 10% (Figure 2.1(A)).
The discounts were more extreme for banks located farther away, as it became more
costly to verify the operational status of those banks, and information asymmetry problem
was exacerbated(Gorton, 1999). Figure 2.1(B) plots the average state bank notes discounts
for four states farther away from Philadelphia. The discounts were as high as 80%, which
meant that a bank note with face value of $1 in Mississippi would only be worth $0.2 in
Philadelphia. The final cost of buying goods in Philadelphia with Mississippi bank notes
would therefore be five times the nominal price. Furthermore, under the unit banking
structure, local price shocks could also easily lead to more bank failures (Bordo, 1998).
The high levels of relative discounts were sometimes due to falling state bond prices
(Rolnick and Weber, 1982) or the anticipation of state bank failures. For example, Illinois
banks committed over 5 million dollars between 1836 and 1842 to building a canal that
would connect the Illinois River and Lake Michigan, hoping to reduce transportation cost
to a larger market. However, this investment completely drained state funds, and caused
a wave of state bank failures in Illinois.19 As a result, the relative discount of Illinois state
bank notes averaged at around 70% in 1842, compared to about 15% in the previous year.
Similarly, Rockoff (1975) estimates that losses on notes due to bank failures ranged from
7% in Indiana to 63% in Minnesota.
The real costs of uncertainty and volatility from circulating multiple currencies created
18 This data is collected in Ales et al. (2008) and the original source is Van Court’s Bank Note Reporter and
Counterfeit Detector published monthly in Philadelphia between February 1839 and December 1858.
19 https://cyberdriveillinois.com/departments/archives/teaching_packages/early_chicago/doc3.html
State bank notes were usually collateralized with state bonds, and significant drops in the value of collateral
often caused bank failures.
72
Figure 2.1: Relative discounts on state bank notes relative to Philadelphia
(a) Regions close to Philadelphia
10
Discount Relative to Philadelphia (%)
-5 0 -10 5
1835m1 1840m1 1845m1 1850m1 1855m1 1860m1

Time
NYC Pennslvania
Massachusetts
(b) Regions far away from Philadelphia

80
Discount Relative to Philadelphia (%)
20 40 0 60
1835m1 1840m1 1845m1 1850m1 1855m1 1860m1

Time
Mississippi Alabama
Louisiana Illinois
Notes: Figure 2.1 plots the monthly average discounts on state bank notes relative to banks in Philadelphia.
Data is from Ales et al. (2008). The original source is Van Court’s Bank Note Reporter and Counterfeit Detector
published monthly in Philadelphia between February 1839 and December 1858.
73
large frictions in exchange and trade, and called attention of policy makers.20 In 1863,
Senator John Sherman from Ohio cited the uncertain values in bank notes as costly
for every citizen. In Congress, he argued for the passage of the National Banking Act
explicitly in terms of securing a stable medium of exchange:
This currency will be uniform. It will be printed by the United States. It will
be of uniform size, shape, and form; so that a bank bill issued in the State of
Maine will be current in California; a bank bill issued in Ohio will be current
wherever our Government currency goes at all; and a bank bill issued in the
State of Connecticut will be freely taken in Iowa or anywhere else. There is
no limit to its convertibility. It will be of uniform value throughout the United
States. I have no doubt these United States notes will, in the end, be taken
as the Bank of England note now is all over the world, as a medium, and a
standard medium of exchange [...] They will be safe; they will be uniform;
they will be convertible. Those are all the requisites that are necessary for any
system of currency or exchange.21
The cost of illiquidity of state bank notes, together with the need to raise money for
the postbellum federal government, eventually led to the passage of the National Banking
Act after the Civil War.
2.2.2 The National Banking Era
The National Banking Act that initially passed in 1863 and was amended in 1864 aimed
to stabilize the banking system and create a network of national banks that were subject
to federal regulations. The newly introduced national banks differed from state banks in
many important ways in this dual-banking system.
First, national bank notes were required by law to have uniform value and be re-
deemable at all national banks in addition to the issuing bank. The bank notes were
backed 110% by federal bonds, which made them risk-free. Compared to the volatility
20 Weber (2003) shows that antebellum interbank payment system was consistent with trade patterns,
which suggests that banks attempted to facilitate trade. See Appendix B.4 for contemporary examples of
how the uncertain value of state bank notes led to legal disputes and inconvenience in exchange. More
recent scholars, such as Cagan (1963), similarly attribute currency stability to economic growth in the
second half of the 19th century.
21 Senate floor, February 10, 1863; http://www.yamaguchy.com/library/spaulding/sherman63.html
74
and uncertainty in the value of state bank notes, the stability and convertibility of national
bank notes made them a better medium of exchange.
Second, national banks were subject to more strict federal regulations that were
designed to prevent bank failures. To limit risk-taking behaviors, national banks were
encouraged to make short-term loans and were not allowed to take land as collateral
(White, 1998).22 These restrictions may have limited capital accumulation that would
require long-term credit, and essentially prevented farmers to obtain credit from national
banks, given their most valuable collateral would be farmland. In contrast, state banks
were often encouraged to extend credit to the agricultural sector, and in fact, some states
even required a minimum fraction of loans to farmers (Knox, 1900). In terms of the
mechanics of regulatory oversight, the Comptroller of the Currency required five reports
on bank operations per year, whereas state banks typically reported their balance sheets
once or twice a year to state regulators.23 The restrictions in banking business and more
rigorous oversight resulted in stability in national banks — between 1875 and 1890, the
average national banks failure rate was about 0.25%, compared to 2.5% of state banks.24
The significantly higher state bank failure rate suggests that state bank liabilities were
risky during the time period we study. Table 2.1 provides a summary of some key
distinctions between national banks and state banks.
National banks also shared some similarities with state banks. Most importantly,
national banks were prevented from branching and operated both privately and locally.
More specifically, any group of 5 or more people could apply for national bank charter
together, and the National Banking Act explicitly required that at 75% of the directors
must have local residency, limiting their ability to choose locations for bank operation.
To induce state banks’ conversion to national banks, the Treasury collected a 2% tax
on state bank notes, and increased it to 10% in 1865. The tax greatly diminished state
22 In contrast, state banks competed with national banks by imposing weaker restrictions on bank
portfolios (White, 1982).
23 See Calomiris and Carlson (2018) for more details on national bank supervision.
24 Data provided by EH.net.
75
Table 2.1: Comparison of National Banks and state banks
State banks National banks

Notes various backing backed 110% by federal bonds
! uniform value
Capital requirement low high
! more costly to establish
Monitoring 2 reports/year to state 5 reports/year to OCC
! more oversight
Stability 2.5% failure rate 0.25% failure rate
! more stable
Lending no restrictions high restrictions
! less lending to farms
Notes: Table 2.1 lists key distinctions between national banks and state banks. Bank failure rates are
calculated for the period between 1875 and 1890.
bank’s ability to issue and circulate their bank notes. However, state bank notes were
still in use for decades.25 To compete with national banks, state banks gradually adopted
checking accounts, which helped them to issue bank debt without being taxed. Checking
accounts provided convenience in transactions, but the uncertainty of their value over
time and in more distant places remained — a check could only be cleared when both
the status of the banks and that of the personal account could be verified. 26 As a result,
national bank notes still had significant advantage in facilitating transactions.
Figure 2.2 shows the evolution of the number and total assets of national banks and
state banks from 1863 to 1900. In the aggregate, national banks replaced many state banks
in the first few years after the Act (Jaremski, 2014). They grew similarly in both number
and total size after 1870, especially between mid-1870s and mid-1880s. We therefore focus
on national bank entries during this steady-state growth period in this paper.
25 Asof the second half of 1870s, many national banks still report state bank notes outstanding.
26 Check clearing also caused more pressure on state banks’ reserves, as when checks are deposited and
cleared the issuing bank would immediately lose reserves, whereas bank notes could be used to settle
transactions without immediate demand for the reserve (Briones and Rockoff, 2005).
76
Figure 2.2: National and state banks: (1863-1900)
(a) Number of banks
(b) Total assets by bank type
Notes: Figure 2.2 plots the total numbers and assets of national and state banks in the United States
between 1863 and 1900. Data is obtained from EH.net operated by the Economic History Association.
77
In view of the historical context, the National Banking Era provides a unique oppor-
tunity to study the impact of financial intermediaries on growth, and it can shed light
on the real costs stemming from unstable bank liabilities. Although few private banks
issue their own currency in the modern day, the fundamental friction of uncertainty in
the value of monetary instruments can be extended to a large variety of assets that are
still used for transactional purposes.
2.3 Data & empirical strategy
In this section, we explain our data sources, sample construction, and the empirical
strategy for identifying national bank entries.
2.3.1 Data sources
In order to study the effects of national bank entry between the mid-1870s to mid-1880s
on outcomes between 1880 and 1890, we combine several historical data sources. Some
examples of the raw sources are shown in Figure 2.3. First, in order to obtain city- and
town-level population, we use the original reports of the 10th and 11th decennial census.
(Figure 2.3(A)) The publicly available digitized census records report total county-level
populations as well as population in areas above 2,500 people, but the city- and town-level
population data are only available in the original census reports. We therefore manually
collected this data on towns and cities.
Second, we use two sources for bank location information. The national bank location
data in 1875 is obtained from the Annual Report of the Comptroller of Currency shown in
Figure 2.3(B). Locations of national banks in 1885 are obtained from The Banker’s Almanac
and Register of 1885. As we can see from Figure 2.3(C), it provides detailed location
information for national banks, as well as state banks and private bankers.
We consider three sets of outcomes. The first set is from the decennial Census of
78
Figure 2.3: Exhibits of data sources
(b) Annual Report of the Office of the

(a) Decennial Census:
Comptroller of Currency (1875)
1880 and 1890
(c) The Banker’s Almanac and Register (1885) (d) Zell’s Business Directory (1875 and 1887)
Notes: Figure 2.3 displays screenshots of data sources that require hand-digitization used in this study.
Figure 2.3(A) shows the town-level population data source, 2.3(B) shows an example from the 1875 Annual
Report of the Comptroller of the Currency, where banks reported their location and balance sheet
conditions. 2.3(C) shows bank location information from the Banker’s Almanac and Register in 1885, and
2.3(D) displays example for local business data.
79
Manufactures and the Census of Agriculture from 1860 to 1900.27 We retrieve output,
input, as well as capital for both sectors in each county, and we obtain per capita measures
by dividing these values by the total number of male laborers above the age of 21. We
focus on per capita measures because county boundaries evolved as new counties were
incorporated throughout the 19th century, and therefore total production values could
be measured incorrectly.28 In addition, we choose adult male laborers as denominator
for two reasons. First, employment by sector was subject to inconsistent reporting both
within and across census years (Carter and Sutch, 1996). Scaling outcomes by the number
of adult male population therefore allows us to better compare production outcomes
between and across sectors and census years. Second, studying outcomes scaled by the
labor force also allows us to better understand the relative magnitudes of the various
components in each sector, such as capital and inputs. One drawback with the census
data is that all values were reported at the county level. In order to better measure the
the effect of national bank entry at the town level, we use the ratio of town population in
our sample to all town population in the county as analytical weight in all regressions
with county-level outcomes.29
Second, we obtain city- and town-level business activities from the Zell’s Classified
United States Business Directory in 1875 and 1887. This directory lists names of all busi-
nesses and professionals in a town (see Figure 2.3(D) for an example), and to the best
of our knowledge, has not been used in prior studies.30 We counted the businesses in
the directory associated with trade-intensive versus not trade-intensive professions for
all towns in 1875 and 1887. Since we lack measures of trade flows between towns, we
27 The Census of Manufacturing does not appear to be available in an easily accessible form for 1910.
The next available census of manufacturing is from 1920, but since that was after both WWI and the
establishment of the Federal Reserve system, we consider it outside the reasonable period of outcomes to
study.
28 For example, (Hornbeck, 2010) adjusts for farmland size changes.
29 The census defined town population as the population residing in towns with above 2,500 people.
30 The 1875 Zell’s Classified United States Business Directory was digitized by the authors from an original
copy in the Boston Public Library. The 1887 directory was obtained from the Baker Business Archives at
Harvard Business School.
80
use the number of commission merchants as a proxy for local trade activity, and use the
number of architects, a profession that was unlikely to be associated with trade activity,
as a placebo occupation. We also complement the Zell’s with occupational records from
the full count censuses of 1880 and 1900.31
Third, we examine the effect on local innovation at the county level using historical
patent data from Petralia et al. (2016). The data provides counts of all patents granted
within counties, which proxy for research and development investments for product
differentiation. Summary statistics for the main variables used in this paper can be found
in Table 2.2.
In some specifications, we also control for the number of state banks in a town as
well as the number of railroads in a county prior to the outcome period. The state banks
location data in 1876 is digitized using The Banker’s Almanac and Register of 1876. Data on
railroad access in 1875 and 1880 is obtained from Atack (2016).
2.3.2 Instrument for national bank entry
Our empirical design relies on the differences in regulatory capital requirements imposed
on national banks based on the size of the town in which the bank was chartered to operate.
We study national bank entry between 1875 and 1885, where the variation in capital
requirements were based on population in 1880.32 We focus on capital requirement
differences based on the 1880 census instead of the 1870 one for two reasons. First,
some towns changed names and incorporation status during the Civil War, causing a
misalignment between 1860 and 1870 census, making it difficult to select towns faced the
same lower capital requirements in the 1860s and study national bank entry post 1870.
Second, in the first few years after the enactment of the National Banking Act, many
31 The 1890 full count census was largely lost in a fire, and the full count 1870 census is currently not
available.
32 The mid-1870s to mid-1880s appears to be a steady-state growth period when the respective growth
trends of national banks and state banks were similar in the aggregate (Figure 2.2).
81
Table 2.2: Summary statistics
Variable Pop1880 2 [4000, 6000] Pop1880 2 (6000, 8000]

Mean Std. Dev. Mean Std. Dev.
Average No. of national banks (1885) 0.4 0.9 0.7 2
Average No. of other financial institutions (1885) 1.1 2.4 1.4 3.2
Average capital (1884, in thousands) 87 48 170 175
Leverage (1884) 3.2 1.0 3.6 1.6
Bank notes per capita (1884) 12.5 10.3 13.6 13.8
Loans per capita (1884) 46.9 35.5 70.3 51.5
Deposit per capita (1884) 37.2 30.3 46.8 31.5
County-level per capita changes (1880:1890):
D Manufacturing production 72.9 163.6 135.6 528.2
D Manufacturing inputs .6 100.6 61.2 445.1
D Manufacturing capital 206.4 179.9 284 308.8
D Agricultural production -25.8 35.1 -23.6 23.3
D Agricultural fixture value -136.5 207.1 -92.70 199.8
D Fertilizers expenditure 0.3 1.0 0.2 1.1
D Commission Merchants (Town-level, 1875-1887) 0 0.52 -0.27 0.93
D Architects(Town-level, 1875-1887) 0.26 0.60 0.41 0.67
% D Buyers and Shippers (County-level, 1880-1900) 0.89 1.19 0.60 0.73
% D Architects(County-level, 1880-1900) 0.75 1.13 0.99 1.68
% D Patents(County-level, 1870s-1880s) 0.53 0.72 1.25 3.12
Number of Observations 126 22
Notes: Table 2.2 presents descriptive statistics of conditions of national bank in the sample as well as main
variables used in the paper. Number of national banks and other financial institutions are obtained from
the Banker’s Almanac and Register of 1885, and bank balance sheet data is from the Annual Report of the
Comptroller of Currency in 1884. County-level per capita changes in manufacturing and agricultural sector
outcomes are calculated from the decennial Census of Manufacturers and Census of Agriculture in 1880
and 1890. Outcomes are scaled by number of male population above age 21. Town-level changes in the
number of commission merchants and architects are calculated from the Zell’s Classified United States
Business Directory in 1875 and 1887. County-level percentage changes in buyers and shippers, and architects
are calculated from the full-count census in 1880 and 1900. Percentage changes the number of patents are
calculated from historical patent data assembled by Petralia et al. (2016).
82
national banks were formed by conversion from state banks (Jaremski, 2014). These state
banks often had large capital stock even though the official requirements were in general
low. As a result, the capital requirement stipulated by population cutoff only had a weak
impact on entries of national banks in the earlier period.
Specifically, the regulation required that:
8
>
>
>
> $50, 000 if population 6,000
>
>
<
Capital stock paid in $100, 000 if 6,000<population50,000
>
>
>
>
>
>
:$250, 000 if population > 50,000.
We focus on the cutoff at 6,000 and use the indicator of a town having below 6,000
population as an instrument for bank entry. 33 It is worth noting that the $50,000
difference in required capital was not trivial for a town in the 1880s: within our sample,
it is about 140 times of average manufacturing wage in 1880. In addition, due to the
no-branching rule, bank owners could not apply for a national bank charter in a small
town but conduct business with customers from a large town. Furthermore, the residency
requirement on bank directors also imposed frictions for towns to seek capital outside of
the town.
Figure 2.4 shows the distribution of town size for all towns with between 2,000 and
10,000 population in 1880, represented by the uncolored bars. The colored bars represent
all towns with fewer than 6,000 people in 1870 that did not have a national bank as of
1875. The insufficient mass in the immediate vicinity of the 6,000 cutoff prevents us from
taking full advantage of a regression discontinuity design. We therefore study towns
within a slightly larger population bandwidth — those with 4,000 to 8,000 population in
the 1880 census (represented by the green bars in Figure 2.4).
Selecting towns with fewer than 6,000 residents in 1870 implies that these towns all
33 Gou (2016), Fulford (2015), and Carlson et al. (2018) also use such variations in capital requirements in
their respective work.
83
Figure 2.4: Histograms of town population in 1880
100
Frequency
50 0
2000 4000 6000 8000 10000

Population in 1880
Sample Outside of [4k,8k]

All
Notes: Figure 2.4 plots the frequency of all towns with 2,000 to 10,000 population in 1880 census (labeled
“All"), as well as after restricting the sample to having below 6,000 population in 1870 and not having a
national bank in 1875 (green and yellow bars). The green bars labeled “Sample" show 1880 population
distribution. All town population data is digitized by the authors from the original census reports.
84
faced the same lower entry cost before the publication of the 1880 census. This allows
us to use differences in bank entry cost after the 1880 census to study subsequent bank
entries and economic outcomes, as some of these towns crossed the 6,000 threshold which
doubled a national bank’s entry cost. As an example, consider Town A and Town B,
each with 4,000 residents as of the 1870 census. In 1880, Town A grew to a population
of 5,000, whereas Town B grew to 7,000 people. Without the capital requirement, the
larger population in Town B would likely cause it to have higher demand for banking
services. However, the capital requirement imposed a bank entry cost on Town B that
was significantly higher than that of Town A.
The location distribution of towns in our sample is shown in Figure 2.5. Figure
2.5(A) separately indicates towns that had populations between 4,000 and 6,000 and those
between 6,000 and 8,000 as of 1880. The map shows that our sample contains more towns
in the northeast, and fewer in the south and the west. Figure 2.5(B) labels the towns that
gained at least one national bank between 1875 and 1885 versus those that did not gain a
national bank during this period.34
2.3.3 Pre-period balance
The identifying assumption for our instrument to be valid is that the towns were similar
in other respects except for likelihood of bank entry. We address this concern in two
ways. First, focusing on a relatively narrow population bandwidth around the 6,000
cutoff partly addresses the concern that the towns were not comparable. Second, we
provide evidence that the average observable characteristics of towns in our sample are
not significantly different as of 1880 except for in population. Panel A of Table 2.3 shows
that the average difference in 1880 population between the two groups of towns is about
2,100, and the difference is mostly driven by larger population growth from 1870 to 1880.
The number of state banks in these towns, number of railroads in 1875 and 1880, as well
34 Figure B2 shows the location distribution of all national banks in 1885.
85
Figure 2.5: Maps of towns in the sample
(a) Towns in the sample
(b) Newly added national banks
Notes: Figure 2.5 maps the locations of towns in the sample. Panel A plots all towns in the sample above
versus below the 6,000 population threshold. Panel B plots the locations of newly added national banks in
our sample.
86
as log market access cost in 1870 (calculated by Donaldson and Hornbeck (2016)) are all
similar for the two sets of towns. Furthermore, average manufacturing and agricultural
production, capital, as well as number of establishments were not different between these
places, as shown in Panels B and C of Table 2.3.
The significant differences in population growth from 1870 to 1880 between the larger
and smaller towns might raise the concern that crossing the 6,000 cutoff in 1880 may
be correlated with later outcomes through non-bank channels. For example, places
that rapidly expanded in the previous decade could continue to grow faster due to
agglomeration effects. We therefore control for population changes between 1870 and
1880 to account for a town’s overall growth trajectory in our analysis as in Carlson et al.
(2018). For robustness, we also control for the number of state banks in a town as of
1876 and the number of railroads in 1875 or 1880 in some specifications. The number
of state banks could proxy for a town’s overall financial conditions before the outcome
period, and the number of railroads could proxy for the area’s overall transportation
development. Both factors could also be important determinants for a place’s future
economic development.
2.3.4 First stage results
We show that our instrument is relevant for national bank entry through the first stage
regression:
(National Bank)i,s = b ⇥ (Pop1880 < 6000)i,s + G0 Xi,s + hs + ei (2.1)
for town i in state s. (National Bank)i,s is an indicator variable for having at least one
national banks in the town as of 1885, and (Pop1880 < 6000)i,s is an indicator variable
for having a town population below 6,000 in 1880 census. Xi,s denotes town characteristics
— population changes between 1870 and 1880, and in some specifications, the number
87
Table 2.3: Pre-period covariate balance
Panel A: Covariate balance
All Pop in [4k,6k] Pop in (6k,8k] Diff
Population in 1870 3970.9 3881.3 4484.1 602.8
(1095.4) (1032.5) (1315.0) (0.02)*
Population in 1880 5059.9 4743.6 6871.5 2128.0
(924.6) (526.9) (541.4) (0.00)***
D Population(1870:1880) 1089.0 862.2 2387.5 1525.2
(1182.6) (1055.7) (1039.8) (0.00)***
No. of state banks in 1876 0.6 0.6 0.8 0.2
(1.0) (1.0) (1.0) (0.49)
No. of railroads (1875) 4.4 4.5 3.9 -0.5
(2.7) (2.6) (3.0) (0.39)
No. of railroads (1880) 5.2 5.3 4.6 -0.7
(2.9) (2.8) (3.2) (0.31)
Log market access cost (1870) 15.8 15.8 15.8 -0.1
(0.5) (0.5) (0.5) (0.53)
N 148 126 22 148
Panel B: Covariate balance (census outcomes in 1880)

Manufacturing, production (1880) 625.5 613.4 694.6 81.1
(458.2) (445.7) (530.7) (0.45)
Manufacturing, capital (1880) 372.4 369.0 391.9 22.9
(278.6) (277.0) (293.8) (0.72)
Manufacturing, establishments (1880) 23.9 24.2 22.2 -2.0
(8.5) (8.2) (10.0) (0.30)
Agriculture, production (1880) 148.4 152.3 126.5 -25.8
(89.8) (91.9) (74.9) (0.22)
Agriculture, capital (1880) 31.7 32.6 26.5 -6.1
(19.2) (19.6) (16.3) (0.17)
Agriculture, establishments (1880) 230.1 232.9 213.8 -19.1
(126.6) (127.5) (122.7) (0.51)
N 148 126 22 148
Panel C: Covariate balance (census outcomes in 1870)

Manufacturing, production (1870) 662.7 666.7 640.3 -26.3
(500.9) (511.4) (446.7) (0.82)
Manufacturing, capital (1870) 347.4 349.9 333.1 -16.8
(273.7) (281.7) (228.1) (0.79)
Manufacturing, establishments (1870) 30.4 31.0 26.8 -4.2
(11.5) (11.1) (13.5) (0.11)
Agriculture, production (1870) 230.4 237.4 191.2 -46.1
(133.7) (135.4) (118.6) (0.14)
Agriculture, capital (1870) 37.3 38.8 28.6 -10.2
(24.9) (25.4) (19.8) (0.08)
Agriculture, establishments (1870) 220.2 222.2 208.7 -13.5
(125.1) (124.0) (133.4) (0.64)
N 147 125 22 147
Notes: Table 2.3 compares the average characteristics of towns above versus below the 6,000 cutoff in or
before 1880. The first column shows the means (and standard deviations in parentheses) for each variable
for All towns in our sample. The second and third columns restrict the sample to places with 4,000 to 6,000
population in 1880 census and 6,000 to 8,000, respectively. The last column shows the differences in means
(and p-values in parentheses) on balance t-tests between the two samples. All values for census outcomes
are per-capita based on the male population above age 21. * p<0.1, ** p<0.05, *** p<0.01.
88
of railroads in 1875 and the number of state banks in 1876. hs denotes state fixed effects.
Results presented in Panel A of Table 2.4 indicate that having population below 6,000 in
1880 is strongly associated with the likelihood of obtaining a national bank in 1885. This
positive relationship is robust to controlling for railroad access and number of state banks
in the towns. The point estimate suggests that the lower regulatory capital requirement
is associated with roughly a 30% higher chance of having a national bank. Although
conversion from state banks to national banks was concentrated in the years between
1863 and 1866, the statistically significant relationship between the number of state banks
in a town in 1876 and having a national bank in 1885 suggests the likelihood that some
national banks in our sample had been converted from state banks.
The relevance of the instrument for national bank entry can be further demonstrated
using falsification tests with alternative population cutoffs near the 6,000 threshold. These
falsification tests address the concerns that national banks tended to establish in smaller
and maybe younger towns due to expectations for higher future growth. We show that
the strong first stage results are unlikely to be induced by factors other than the capital
requirements in the National Banking Act, as indicated by results presented in Panel B of
Table 2.4.
Next, we show that pre-period observable characteristics were not sorted by the
population cutoff at 6,000, conditional on population growth and state fixed effects.
Figure 2.6 plots unstandardized coefficients from the regression
Yi,s = b ⇥ (Pop1880 < 6000)i,s + Dpopi,s + hs + ei , (2.2)
where Yi,s denotes various county-level outcomes from the Census of Manufacturers
and Census of Agriculture in the periods 1870 and 1880, railroad access in 1875 and
1880, market access cost in 1870, as well as the town-level number of state banks in
1876. The number of state banks in 1876 and railroad access in 1875 and 1880 could
89
Table 2.4: First stage regressions
Panel A: First stage: preferred cutoff (6k)
1(National Bank)
(1) (2)
1(pop<6k) 0.305*** 0.286***
(0.104) (0.0997)
D Pop(1870:1880) 0.000135*** 0.000133***
(0.0000354) (0.0000341)
No. of railroads (1875) -0.00349
(0.0159)
No. of state banks (1876) 0.137***
(0.0398)
State FE Y Y
Adj. R2 0.229 0.289
N 148 148
Panel B: First stage: alternative cutoffs
(1) (2) (3) (4) (5)

5000 5500 6000 6500 7000
Population Cutoff 0.0646 0.103 0.305*** 0.219* 0.209
(0.0729) (0.0922) (0.104) (0.113) (0.147)
D Pop(1870:1880) 0.0000964*** 0.000104*** 0.000135*** 0.000107*** 0.000100***
(0.0000348) (0.0000364) (0.0000354) (0.0000336) (0.0000336)
State FE Y Y Y Y Y
Adj. R2 0.180 0.183 0.229 0.199 0.188
N 148 148 148 148 148
Notes: Panel A of Table 2.4 presents results from the first stage regressions:
xt = a + b ⇤ 1(Pop1880 < 6000) + Zt + et where xt is either an indicator of gaining at least one national
bank, or the number of national banks. Panel B presents results of the first stage regression using indicator
of having at least one national banks in 1885 as dependent variable, and various population cutoffs as the
RHS variable. * p<0.1, ** p<0.05, *** p<0.01.
be considered proxies of general conditions of a place’s financial development and

market access, which could influence growth potential in later decades. Similarly to
Table 2.3, towns that were above or below the 6,000 cutoff did not have significant
differences in all of these characteristics. Moreover, changes in per capita manufacturing
and agriculture production, capital and number of establishments from 1870 to 1880 were
90
also not significantly different. Similarity in all these observable characteristics provides
reassuring evidence that our instrumental variable constructed from population cutoff
does not correlate with local economic conditions prior to national bank entries.
Figure 2.6: Conditional covariates balance
Manu. Prod. per capita (1880) Manu. Capital per capita (1880)
Agri. Prod. per capita (1880) Agri. Capital per capita (1880)
Manu. Prod. per capita (1870) Manu. Capital per capita (1870)
Agri. Prod. per capita (1870) Agri. Capital per capita (1870)
Δ Manu. Prod. per capita (1870:1880) Δ Manu. Capital per capita (1870:1880)
Δ Agri. Prod. per capita (1870:1880) Δ Agri. Capital per capita (1870:1880)
-300 -200 -100 0 100 -150 -100 -50 0 50 100
Number of Manu. per capita (1880) Number of Railroads (1875)
Number of Farms per capita (1880) Number of Railroads (1880)
Number of Manu per capita (1870) Number of State Banks (1876)
Number of Farms per capita (1870) Log Market Access Cost (1870)
Δ Number of Manu per capita (1870:1880) 1(National Bank)
Δ Number of Farms per capita (1870:1880) -2 -1 0 1 2
-.05 0 .05 .1 .15 99 95 90
Notes: Figure 2.6 displays unstandardized coefficients on the indicator of having below 6,000 population in
1880 for various outcomes Yi,s (production, capital and number of establishments/farms) from the
regression Yi,s = a + b1( pop < 6, 000)i,s + Dpopi,s + gs + ei,s for town i in state s. The darkest shades
represent 90% confidence intervals and the lightest shades represent 99% confidence intervals.
The relationship between the population cutoff at 6,000 and bank entry remains robust
in wider population ranges around the 6,000 cutoff, and we provide more details in
Appendix B.3. We choose the more conservative sample size for our main results since a
greater population window raises concerns about comparability between the larger and
smaller places. In addition, our first stage results are also robust to selecting a base year
before the 1880 census and end-year after the 1880 census, and in Table B1 we provide two
examples for periods between 1873 and 1883, and between 1877 and 1887. Furthermore,
we present evidence in Figure B3 that national bank entries did not cluster in years right
before the 1880 census was published, both in our sample and in the aggregate. In sum,
91
national banks did not appear to have entered to a towns due to anticipation of their
higher future growth, but bank entries were significantly affected by the population-based
instrument.
2.4 Results
In this section, we present the results on how access to national banks impacted the
local economy. We start by studying changes in the agricultural sector, as national banks
provided limited lending to farmers due to their inability to extend loans collateralized
by farmland. This condition provides us an ideal setting to fully explore the impact of
stable bank liabilities on local economic development.
2.4.1 Changes in the agriculture sector
Agriculture was a growing sector in the late 19th century, especially in the west. On
one hand, the rapid expansion of the railroad network reduced transportation costs of
agricultural products (Donaldson and Hornbeck, 2016); on the other hand, national banks
provided limited credit to farmers due to their inability to take farmland as loan collateral.
This lending restriction likely limited farmers’ ability to acquire new farmland or improve
their current property (Fulford, 2015; Knox, 1900).
We focus on the one-period difference in county-level agricultural outcomes per capita
between 1880 and 1890 and estimate:
DYi,s = be(National Bank) + G0 Xi,s + hs + # i,s (2.3)
for town i in state s. e(National Bank) is an indicator of having at least one national
bank in 1885, instrumented by the indicator of a town’s population being below the
cutoff at 6,000. b is the main coefficient of interest, which measures the change of the
output response to having a national bank. Xis is a vector of control variables, and hs
92
denotes state fixed effects, which capture certain state-level characteristics that could affect
manufacturing production growth such as state tax or subsidy for agricultural production.
Time-invariant characteristics of the counties are subsumed by the differences. The
outcome variables of interest are calculated as per capita based on the male population
above the age of 21.
We study three outcomes — agricultural outputs, value of land and fixtures (fences
and buildings), as well as expenditure on fertilizers per capita. Changes in agricultural
outputs are useful in gauging national banks’ overall impacts on the agriculture sector.
To better disentangle the various effects of national banks, we use changes in the value of
farmland and fixtures as a proxy for the access to long-term credit, which was usually
granted over a longer period on mortgage security (Pope, 1914). In contrast, credit for
expenditure on fertilizers is considered as “working capital," as fertilizers were common
inputs in the agricultural production. Therefore, fertilizers expenditure is a proxy for
short-term credit accessibility. One additional advantage to study changes in fertilizers
expenditure is that commercial fertilizer was first introduced in the 1840s, which was
long before our sample period. Therefore, differences in the adoption of fertilizers was
unlikely due to factors such as access to new innovation or information.
We find that none of agricultural production, farm’s land and fixture value, or
fertilizer expenditure was affected by national banks, as shown in Table 2.5. The results
are consistent with national bank’s limited provision of long-term credit for agricultural
expansion and improvement on mortgage security. The insignificant results on changes
in fertilizer expenditure further indicate that national banks did not play a significant role
in short-term credit provision for agricultural inputs either. One possible explanation is
that farmers did not have strong relationships with national banks because of the lending
restrictions, and it also prevented them from taking out short-term loans from national
banks.
Although total agricultural production was unaffected by entries of national banks, we
93
Table 2.5: National Banks and local agricultural sector (1880-1890)
Panel A: Changes in agricultural production outputs per capita
OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) -11.36 -3.316 -9.331 3.708
(7.140) (7.816) (19.33) (24.90)
1(pop<6k) -4.097 1.253
(9.396) (9.280)
State FE Y Y Y Y Y Y
Controls N Y N Y N Y
Mean of Dep. Var. -25.48 -25.48 -25.48 -25.48 -25.48 -25.48
Std. Dev. of Dep. Var. 33.54 33.54 33.54 33.54 33.54 33.54
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Panel B: Changes in farmland value per capita
OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 12.46 -4.666 37.25 32.21
(32.40) (35.33) (87.91) (112.7)
1(pop<6k) 16.36 10.89
(42.25) (41.91)
Mean of Dep. Var. -130.0 -130.0 -130.0 -130.0 -130.0 -130.0
Std. Dev. of Dep. Var. 205.9 205.9 205.9 205.9 205.9 205.9
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Panel C: Changes in expenditure on fertilizers per capita
OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 0.269 0.0279 0.185 -0.164
(0.243) (0.270) (0.659) (0.860)
1(pop<6k) 0.0812 -0.0554
(0.319) (0.321)
Mean of Dep. Var. 0.276 0.276 0.276 0.276 0.276 0.276
Std. Dev. of Dep. Var. 1.030 1.030 1.030 1.030 1.030 1.030
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Notes: Table 2.5 presents results from OLS, reduced form, as well as IV estimates. Dependent variable is
changes in agricultural production, value of farmland and fixtures (fences and buildings), and expenditure
on fertilizers per capita between 1880 and 1890. Control variables include number of state banks in a town
as of 1876, as well as number of railroads in 1880. Regressions are weighted by share of town population in
the sample in 1880.* p<0.1, ** p<0.05, *** p<0.01.
94
show in Table 2.6 that production shifted significantly from non-traded crops to traded
crops. “Traded crops" are defined as crops that were exchanged on the Chicago Board
of Trade, which include wheat, oats, buckwheat, and Indian corn. In 1880, non-traded
crops were only about 9.4% of total production, because American agricultural products
exports increased greatly in the last quarter of the 19th century (Pope, 1914). The point
estimates in Column 5 and 6 mean that 75% or all of production shifted from non-traded
crops towards traded crops.
Table 2.6: National Banks and the shift in agricultural production
OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) -0.0167* -0.0247** -0.0710*** -0.102***
(0.00856) (0.00949) (0.0267) (0.0375)
1(pop<6k) -0.0312*** -0.0345***
(0.0110) (0.0111)
Mean of Fractions in 1880 0.0943 0.0943 0.0943 0.0943 0.0943 0.0943
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Notes: Table 2.6 presents results from the OLS, reduced form, and IV estimates of the effects of national
banks on changes in the fractions of non-traded crops to total crops measured in bushels between 1880 and
1890. All columns include town population changes between 1870 and 1880 as a control variable.
Additional control variables include the number of state banks in town as of 1876 and number of railroads
as of 1880. Regressions are weighted by share of town population in the sample in 1880. * p<0.1, ** p<0.05,
*** p<0.01.
While we cannot completely rule out the possibility that traded crops were more
capital-intensive in some ways, the insignificant impact of national bank entries on both
long-term and short-term credit obtained by farmers indicates that bank lending played
little role in the agriculture sector in our sample overall.35 The results provide evidence
that national banks may have impacted production decisions through providing more
35 Fulford(2015) finds that national banks positively affected agricultural sector production through
providing working capital and liquidity in rural counties. Our samples may have little overlap as we use
town-level population instead of county-level population for identification.
95
secure bank liabilities that facilitated transactions and trade. The shift towards traded
crops production is also consistent with our finding on increased trade activity following
national banks’ entry.
2.4.2 Trade activity
We provide further evidence on the positive effect of national bank entry on local trade
activity with both town- and county-level outcomes in this subsection.
Growth in trade-related occupations
We capture inter-regional trade flows by proxying it with trade-related occupations.36

Commission merchant was a profession that sourced products from one place and sold at
another. The merchants did not directly participate in production, but simply profited by
facilitating trade and sales. Therefore, the number of commission merchants in a town
can serve as an indicator of the volume of products sourced from potentially distant areas.
We examine changes in town-level number of commission merchants, collected from the
Zell’s Classified United States Business Directory in 1875 and 1887.
We estimate the baseline Equation 2.3 to study the effect of gaining national banks on
local trade activity by replacing the outcome variable to the changes in the number of
commission merchants in a town between 1875 and 1887. Results are reported in Table
2.7. The IV estimates in Columns 5 and 6 of Panel A suggest that gaining a national bank
between 1875 and 1885 is associated with 1.3 to 1.5 more commission merchants within
a town in 1887. The effect is large compared to the mean of 0.14 in 1875. As a placebo
test, we also analyze the relationship between national bank entries and changes in the
number of architects — a profession that was likely to operate locally. we find no effect
of national banks in all specifications, as shown in Table 2.7 Panel B.
To complement our analysis with the business directory data, we also use the full
36 There was no official domestic trade flow statistics in the 19th century United States.
96
Table 2.7: National Banks and local business (town-level)
Panel A: Changes in the number of commission merchants
OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 0.0624 0.189 1.278** 1.447**
(0.137) (0.141) (0.624) (0.650)
1(pop<6k) 0.390** 0.413***
(0.159) (0.156)
Average No. in 1875 0.135 0.135 0.135 0.135 0.135 0.135
N 148 148 148 148 148 148
KP F-stat 8.676 8.212
Panel B: Changes in the number of architects (placebo test)
OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 0.0126 0.0302 0.193 0.201
(0.119) (0.125) (0.427) (0.453)
1(pop<6k) 0.0590 0.0573
(0.142) (0.142)
Average No. in 1875 0.00676 0.00676 0.00676 0.00676 0.00676 0.00676
N 148 148 148 148 148 148
KP F-stat 8.676 8.212
Notes: Table 2.7 presents results from the OLS, reduced form, and IV estimates of the effect of national
banks entries on changes in number of trade-related business (commission merchants) and local-oriented
business (architects). Changes are from 1875 to 1887. All columns include town population changes
between 1870 and 1880 as a control variable. Additional control variables include the number of state
banks in town as of 1876 and number of railroads as of 1875. Regressions are equal-weighted. * p<0.1, **
p<0.05, *** p<0.01.
97
count census of 1880 and 1900 which contains more occupation categories. We contrast
growth in trade-related workers (buyers and shippers) to occupations that were unlikely
to be affected by trade (architects, doctors, and teachers).37 As before, we scale the number
of workers in these occupations by county-level male population above the age of 21. The
outcome variables are the growth rates in shares of the occupations as outcome variables.
We find that gaining a national bank positively impacted trade activity at the county
level. Gaining a national bank led to 1.6 to 1.7 times higher changes in the share of buyers
and shippers between 1880 and 1900, as shown in Panel A of Table 2.8. We again find
no significant impact of national banks on changes in the share of architects as shown
in Panel B of Table 2.8. More placebo test results using growth in shares of doctors and
teachers can be found in Table B3.
Both town-level and county-level results show that gaining access to national banks
led to more trade-related activity. The evidence is consistent with national bank’s ability
to provide more secure bank liabilities that could facilitate transactions with distant
counterparties. The lower transactions frictions with national bank currencies may have
propelled manufacturing sector growth by eliminating nominal price risks in sourcing
inputs from and selling outputs to more locations, and providing local manufacturers
greater access to inputs and outputs markets. However, we do not rule out national banks’
impact on trade activity through the traditional lending channels. In fact, the significant
greater change in manufacturing inputs per capita following national bank entry provides
evidence that national banks likely provided short-term credit to the manufacturers. The
short-term credit could also have been used for trade finance, which would also lead to
increased local trade activity.
37 The full count census of 1890 was lost in a fire.
98
Table 2.8: National Banks and local business (county-level)
Panel A: Percentage changes in share of buyers and shippers
OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) -0.213 -0.158 1.583* 1.727*
(0.261) (0.274) (0.835) (0.883)
1(pop<6k) 0.556** 0.574**
(0.269) (0.270)
Mean of Dep. Var. 0.850 0.850 0.850 0.850 0.850 0.850
Std. Dev. of Dep. Var. 1.139 1.139 1.139 1.139 1.139 1.139
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Panel B: Percentage changes in share of architects (placebo)

OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) -0.0157 0.123 -0.548 -0.478
(0.305) (0.314) (0.839) (0.870)
1(pop<6k) -0.192 -0.159
(0.319) (0.315)
Mean of Dep. Var. 0.788 0.788 0.788 0.788 0.788 0.788
Std. Dev. of Dep. Var. 1.221 1.221 1.221 1.221 1.221 1.221
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Notes: Table 2.8 presents results from the OLS, reduced form, and IV estimates. Dependent variables in
Panel A and B are percentage changes in shares of buyers and shippers, and architects, among males above
21 years old from 1880 census to 1900 census, respectively. All columns include town population changes
between 1870 and 1880 as a control variable. Additional control variables include the number of state
banks in town as of 1876 and number of railroads as of 1880. Regressions are weighted by share of town
population in the sample in 1880. * p<0.1, ** p<0.05, *** p<0.01.
99
Complementary evidence from prices
We also provide some complementary evidence on how national banks may have reduced
transactions costs by comparing price changes in “trade-sensitive" goods versus “local"
goods following national bank entry. Sellers of traded products had to bear the price
risk associated with the uncertain currency values between their towns and the towns
where they sourced the products. Therefore, price uncertainty was likely to drive up
costs, leading to higher sale prices locally. On the other hand, selling locally produced
goods did not involve transactions with non-local bank notes.
We collected data on the price of tea, New Orleans molasses, and starch from 1864 to
1880 in 9 towns from the supplementary reports on The Average Retail Prices of Necessaries
of Life in Statistics of Wages published by the census office in 1886. These 9 towns had their
first national banks between 1866 and 1878. We categorize tea and New Orleans molasses
as “trade-sensitive" goods, as they were either imported and distributed from the ports,
or produced specifically in New Orleans. Starch, on the other hands, is categorized as
“local" good as it was likely to be locally produced from corn.
We find that the price of tea and New Orleans molasses dropped significantly with
the access to national banks, whereas the price of starch was not impacted. As shown in
Table 2.9, the price of tea dropped by about 30 cents after the towns had national banks,
relative to the average price of $1.2 per pound (a 25% drop ). Similarly, the price of New
Orleans molasses dropped by about 25 cents per gallon from $1.1 per gallon (a 23% drop).
The results suggest that national banks may have helped to reduce transaction cost by
providing a stable medium of exchange, and therefore positively impacted trade-intensive
economic activity.
100
Table 2.9: Price levels and existence of National Banks
(1) (2) (3)

Price of Tea Price of Molasses Price of Starch
1(National Bank) -0.294*** -0.250*** -0.0105
(0.0844) (0.0651) (0.00775)
Year FE Y Y Y
Pre-NB Mean 1.238 1.120 0.120
N 115 105 117
Adj. R2 0.229 0.350 0.147
Notes: Table 2.9 presents results from estimating Pit = a + b1(National Bank)it + gt + # it for tea, New
Orleans molasses, and starch in 9 towns over 1864 to 1880. The first years of having at least one national
bank in these town range from 1866 to 1878. Standard errors are clustered by year. *** p<0.01, ** p<0.05, *
p<0.1.
2.4.3 Growth in the manufacturing sector
Having shown that national bank entry led to increased local trade activity, we turn to
study the effect of national banks on the production in the manufacturing sector, since
manufacturing outputs are considered as tradable products. We examine whether the
gaining a national bank also led to higher manufacturing production as well as the
possible driving forces.
We find that national bank entry led to economically and statistically significant higher
growth in manufacturing production per capita between 1880 and 1890 even though we do
not focus on the “manufacturing belt" states as in Jaremski (2014) and Carlson et al. (2018).
We estimate the main specification in Equation 2.3 using differences in manufacturing
production per capita as the outcome variable. The IV estimates from Table 2.10 show
that gaining a national bank led to about $229 to $310 higher growth in manufacturing
production per capita, which are about $6,360 to $8,620 in 2018 dollars. The average effect
is 50% of the average of pre-period levels in 1880, which is very economically significant.
The smaller and statistically insignificant OLS estimates reported in Column 1-2 in
Panel A indicate that national banks did not selectively enter into high-growth places.
101
Table 2.10: National Banks and growth in manufacturing production
Panel A: Changes in manufacturing production per capita

OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 39.87 57.43 228.5** 310.0**
(32.97) (36.16) (100.4) (136.0)
1(pop<6k) 100.4** 104.8**
(42.28) (42.29)
Mean of Dep. Var. 82.21 82.21 82.21 82.21 82.21 82.21
Std. Dev. of Dep. Var. 251.2 251.2 251.2 251.2 251.2 251.2
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Panel B: Changes in manufacturing production per capita — subsamples
OLS RF IV OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 56.13 312.0** 76.35* 386.3**
(39.56) (123.2) (38.64) (171.1)
1(pop<6k) 140.0*** 114.9**
(50.78) (44.38)
Same Treatment in County Y Y Y
Excluding CA and PA Y Y Y
Mean of Dep. Var. 98.33 98.33 98.33 80.44 80.44 80.44
Std. Dev. of Dep. Var. 260.4 260.4 260.4 258.6 258.6 258.6
N 123 123 123 139 139 139
KP F-stat 12.55 7.987
changes in manufacturing production per capita between 1880 and 1890. All columns include town
population changes between 1870 and 1880 as a control variable. Additional control variables include the
number of state banks in town as of 1876 and number of railroads as of 1880. Regressions are weighted by
share of town population in the sample in 1880. In Panel B, Column 1-3 report results on counties where
either all towns in sample are above or below the 6,000 cutoff. Column 4-6 report results on all states
except for California and Pennsylvania. * p<0.1, ** p<0.05, *** p<0.01.
102
There may be two reasons why this was the case. First, location selection was greatly
limited by the residence requirement for national bank’s directors. The National Banking
Act required that a national bank must have at least 5 directors, and at least 75% of them
must have resided locally. This residence requirement generated frictions in selecting
locations for bank operation, such that places that would likely benefit the most from
national banks may not have suitable directors for them. Second, even in this later period,
some national banks may have been converted from state banks. Since state banks were
known to lend more to farmers (Knox, 1900), places with more state banks previously
might have had a relatively underdeveloped manufacturing sector. As a result, there
would be no significant effect of national banks on subsequent manufacturing growth
overall.
Having established the main effects on manufacturing production per capita, we
repeat the same analysis on two different subsamples to demonstrate robustness. The first
subsample excludes all counties that had towns both above and below the 6,000 cutoff,
and the second subsample excludes California and Pennsylvania. Results are shown in
Panel B of Table 2.10.
The purpose for using the first subsample is to facilitate interpretation of the magni-
tude of the point estimates. There are a few counties containing both towns above and
below the 6,000 cutoff, and using per capita manufacturing production calculated with
county-level data could be less interpretable for these counties. As shown in Columns
1-3, after excluding these places, we still find a significant increase in manufacturing
production per capita of similar magnitude to the baseline results.
The second subsample addresses concerns that changes in town size and growth were
both due to gold rushes and discoveries of coal mines in each state, respectively. Gold
rushes and new discoveries of coal mines were unlikely an important factor in our results.
There were only a few gold rushes in the late 19th century United States compared to
earlier of the century, and the coal industry was already relatively mature. Between 1870
103
and 1890, only the late 1870s Bodie Gold Rush in California (Sprague, 2011) and discovery
of coking coal in the Connellsville region of southwestern Pennsylvania38 could have
impacted places in our sample. Results from Column 4-6 suggest that excluding towns in
California and Pennsylvania has little impact on the main results.
We report more robustness and placebo test results in Appendix B.1. Results reported
in Table B4 show that our main results are robust to alternative definition of labor force
and inflation adjustment. In Panel A of Table B4, we calculate manufacturing production
per capita by scaling production by male population between 18-44 years old, or the
“prime-age" male labor. In Panel B, we adjust 1880 production value to 1890 dollars.
Reliable CPI series was only published after the establishment of the federal reserve
system in 1913, however, based on a limited price index, the Bureau of Labor Statistics
estimates that there was about 10% deflation between 1880 and 1890.39
We also conduct a set of pre-period placebo tests using changes in manufacturing
production per capita from 1870 to 1880 as the outcome variable, and results are reported
in Table B5. We find no evidence that gaining a national bank between 1875 and 1885 had
an impact on manufacturing growth in the 1870s, which further rules out the possibility
of pre-trend in growth having simultaneous effects on town size and subsequent growth.
Decomposing growth in the manufacturing sector
Given the significant effect on manufacturing production, in this subsection we examine

the components of manufacturing growth, including inputs, employment, and capital
between 1880 and 1890. We re-estimate the baseline specification in Equation 2.3 with the
different components of manufacturing sector as the dependent variable, and present the
results in Table 2.11.
Manufacturing inputs value increased significantly more after gaining access to na-
tional banks. The IV estimates in Columns 5 and 6 of Panel A indicate that gaining a
38 Source: https://eh.net/encyclopedia/the-us-coal-industry-in-the-nineteenth-century-2/
39 Source: https://www.officialdata.org/
104
Table 2.11: National Banks and growth in manufacturing inputs, employment, and capital
Panel A: Changes in manufacturing inputs per capita
OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 27.01 34.75* 153.5*** 205.4**
(18.41) (20.53) (58.61) (81.69)
1(pop<6k) 67.40*** 69.40***
(23.43) (23.83)
Mean of Dep. Var. 9.643 9.643 9.643 9.643 9.643 9.643
Std. Dev. of Dep. Var. 193.3 193.3 193.3 193.3 193.3 193.3
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Panel B: Changes in manufacturing employment per capita

OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 0.0194 0.0351* 0.113** 0.164**
(0.0188) (0.0209) (0.0557) (0.0760)
1(pop<6k) 0.0494** 0.0555**
(0.0242) (0.0245)
Mean of Dep. Var. 0.0740 0.0740 0.0740 0.0740 0.0740 0.0740
Std. Dev. of Dep. Var. 0.0868 0.0868 0.0868 0.0868 0.0868 0.0868
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Panel C: Changes in manufacturing capital per capita

OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 2.508 44.01 55.86 132.0
(37.18) (41.16) (101.5) (133.1)
1(pop<6k) 24.53 44.62
(48.42) (48.90)
Mean of Dep. Var. 218.0 218.0 218.0 218.0 218.0 218.0
Std. Dev. of Dep. Var. 204.7 204.7 204.7 204.7 204.7 204.7
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
changes in manufacturing inputs, employment, and capital per capita between 1880 and 1890. All columns
include town population changes between 1870 and 1880 as a control variable. Additional control variables
include the number of state banks in town as of 1876 and number of railroads as of 1880. Regressions are
weighted by share of town population in the sample in 1880.* p<0.1, ** p<0.05, *** p<0.01.
105
national bank led to $154 - $205 more in manufacturing inputs per capita, which are about
$4,280 and $5,700 in 2018 dollars or 54% of the average level in 1880. The magnitudes
here suggest that the increase in inputs value was responsible for 66% of the total output
increase, assuming constant returns to scale. Similarly, gaining a national bank also led to
significant growth in manufacturing employment. The IV estimates in Columns 5 and 6 of
Panel B show that about 11-16 more people per 100 males above age 21 were employed in
the manufacturing sector following national bank entry. However, the impact of national
banks on manufacturing capital was positive but statistically insignificant, as shown in
Panel C of Table 2.11. The magnitude ($56-$132) is also smaller compared to effects of
national banks on manufacturing inputs.40
The results above on the decomposition of manufacturing output indicate that national
banks led to significant growth through increased inputs and employment, but that their
impact on capital investment was limited. There are several possible explanations for
the null effect on capital. First, regulations encouraged national banks to make short-
term loans rather than long-term loans (White, 1998). These loans provided working
capital to meet short-term liquidity needs rather than long-term investment demands.
The short-term credit may have also facilitated inputs sourcing, and the lack of long-
term credit provision limited manufacturers’ ability to acquire physical capital such as
manufacturing factory land. Second, since the national banks were unit banks, their
ability to diversify their loan portfolio with borrowers across different places was limited,
which could discourage them from expanding their balance sheets. Third, large values
of firm investment could not be easily accommodated, as a national bank could lend no
more than 10% of its capital stock to one entity.41 Broadly, many requirements impeded
investment in the manufacturing sector, suggesting that the effect of national banks from
the asset side of the balance sheet is limited.
40 Carlson et al. (2018) find positive effect on manufacturing capital growth following new national bank
entry, and suggest that banking competition leads to economic growth by inducing credit provision.
41 That is to say, a bank with $50,000 of capital stock could lend no more than $5,000 to a firm.Source:
National Banking Act of 1864, Sec. 29
106
One caveat to our interpretation is that manufacturing capital was estimated by
establishment owners and was subject to mis-measurement in the census, which would
bias the results toward zero. Remarks on the Statistics of Manufactures for the 1880 census
explains how factory owners may not have correctly estimated the value of all physical
capital used in production. For example, a manufacturer who rented a piece of factory
land may or may not count the land as a part of capital. The measurement error is
therefore likely to weaken the relationship between national banks and manufacturing
capital growth.
So far, the results indicate that national banks facilitated manufacturing production
from the assets side primarily by providing working capital. Given the large magnitude
of impacts of national banks on manufacturing outputs and inputs, we explore other
channels that could also have spurred the growth in the manufacturing sector. In Section
3.2, we provided an overview of the significant exchange frictions stemming from illiquid
state bank notes. Since the National Banking Act required that the national bank notes
must be redeemable at all national banks at par, we hypothesize that places with access
to national bank notes experienced significant reduction in transaction cost when trading
with distant regions. As a result, value of manufacturing production and inputs could
both increase with greater access to inputs and outputs markets. We provide evidence for
this channel next.
2.4.4 Growth in innovation
As the late 19th century was a part of the Second Industrial Revolution, when organized
industrial R&D within firms emerged and science and technology were applied to
product development (Bruland and Mowery, 2006), we postulate that the significant
manufacturing production increase could also be driven by innovation. We examine
whether national banks also had a significant impact on local innovation activity in this
subsection.
107
We use the the number of patents obtained by residents within a county as measure
for innovation activity (Petralia et al., 2016).42 . We first show that the number of patents
can be a reasonable proxy for the manufacturing sector’s research and development
outcome by plotting the relationship between manufacturing and agricultural production
per capita and the log number of patents obtained by local inventors in the previous
decade in Figure 2.7. The plots show that local innovation output measured by the
number of patents granted is strongly correlated with manufacturing production per
capita, but is only weakly correlated with agriculture production per capita. Assuming
patents contributed to total factor productivity, these correlations indicate that they were
primarily innovations that benefited the manufacturing sector.
Having established that patents are a suitable proxy for manufacturing sector’s
innovation, we examine the percentage changes in the number of patents granted from
the decade following 1870 and 1880 using the specification of Equation 2.3.43 Gaining
a national bank led to 105 to 134 percentage points higher increase in the number of
patents across the two decades, as shown in Table ??. The magnitude is economically
significant, compared to the mean of 64 percentage points and standard deviation of
138 percentage points. We also conduct placebo tests by replacing the outcome variable
with percentage changes in number of patents between the previous two decades (1860s
and 1870s) and report the results in Table B6. The insignificant results imply that the
difference in innovation output growth was not due to pre-existing trends.
Assuming that firms primarily relied on internal financing for R&D like they do in
the modern period in the United States (Bougheas, 2004; Hall, 2002), the results also
indicate that the positive effect of national banks on innovation was not through the
bank lending channel. Instead, the evidence is consistent with that manufacturers gained
greater market access and were exposed to a greater varieties of products when product
price uncertainty across locations is reduced after gaining a national bank locally. As
42 Plant- or sector-level innovation measure is not available for our sample period
43 We include the 10 years between 1871 to 1880, and 1881 to 1890, respectively.
108
Figure 2.7: Production per capita and innovation
(a) Manufacturing production per capita and log number of patents
400
Manufacturing production per capita
100 2000 300
0 1 2 3 4
Log number of patents in the previous decade
(b) Agriculture production per capita and log number of patents

300
Agriculture production per capita
200 150250
0 1 2 3 4
Log number of patents in the previous decade
Notes: Figure 2.7 presents binscatter plots between manufacturing/agriculture production per capita in
1860-1900 census and log number of patents per county in the previous decade. The plots control for total
county population, as well as state-year fixed effects.
109
a result, manufacturers may have incentive to to innovate and therefore become more
competitive in a larger market and expand product lines.
The increased trade and innovation activity could both help to explain the significant
growth in manufacturing production after national bank entries. Taken together, national
banks likely facilitated manufacturing sector growth from both the assets and the liabilities
sides of their balance sheets. On the assets side, short-term loans may have provided the
working capital and liquidity for input sourcing and trade finance. On the liabilities side,
granting easy access to secure and liquid bank liabilities may have increased production
value by reducing transactions cost in trade and hence allowing local manufacturers to
access more customers and produce more differentiated goods.
2.4.5 Long-term effects and spillover effects
In this part, we study whether the positive impacts of national banks on manufacturing
production persisted over time led to any impacts on neighboring counties. The initial
significant growth in the manufacturing sector may last for a longer time period due to
the presence of agglomeration economies in manufacturing (Kline and Moretti, 2014b).
In addition, greater industrialization, greater market access through trade and more inno-
vation activity could also turn into comparative advantages in manufacturing production,
which may sustain higher manufacturing production over a longer period (Sequeira et al.,
2018). We test this hypothesis next.
Long-term effects of national banks
To measure the long-term effects of national banks, we estimate a dynamic differences-in-

differences model from 1860 to 1900, which makes it possible to visualize any pre-trends
before the 1880 census was published and the persistence afterward. Specifically, we
110
estimate the following:
Yist = Â bk e(National Bank) ⇥ {year =k} + Â gk X 0 ⇥ {year =k} + hst + ai + # it , (2.4)

k k
where Yist is the outcome in town i in state s at year t. We include leads and lags before
and after 1880, and omit 1880 so that all outcomes are relative to the treatment period.
b k is the coefficient of interest, and it measures the elasticity of the output response to
gaining a national bank in each of the lead and lag years. As before, e(National Bank) is
the instrumented variable. gk allow the control variables X 0 to have time-varying effects.
hst are state-year fixed effects so that we compare outcomes within states and years, and ai
are town fixed effects that control for time-invariant characteristics such as geographical
location and land quality.
The full IV and reduced form dynamic difference-in-difference coefficients for manu-
facturing production plotted in Figure 2.8 show that the positive effect of national banks
on manufacturing production between 1880 and 1890 persisted beyond a decade into
1890s as well. The figures also indicate that there were no differential pre-trends between
the places that received national banks versus those that did not, instrumented by the
population cutoff at 6,000.44 Data from the 1910 Census of Manufacturers is not available,
and we find no differential effect using the 1920 outcomes. However, 1920 was after the
WWI and the establishment of the federal reserve system. We therefore do not attribute
the result to conversion of growth.
Spillover effects
We also investigate whether there are indirect effects of national bank entries on the
neighboring counties. On one hand, more active trade and increased innovation output
could create positive spillover effects on the neighboring counties; on the other hand,
44 We only show 2 pre-shock period coefficients here, as 1840 and 1850 census does not change the flat
patters of the pre-shock trend.
111
Figure 2.8: Persistent positive effect on manufacturing production
1500
manufacturing prod per capita
0 500-500 1000
1860 1870 1880 1890 1900
Reduced form IV
N = 733
Notes: Figure 2.8 shows the dynamic diff-in-diff coefficients for the reduced form and the IV estimates of
the effect of having a national bank on county-level manufacturing production value. The specification for
the IV estimates is Yist = Âk b k e (National Bank) ⇥ {year=k} + Âk gk X 0 ⇥ {year=k} + hst + ai + # it . 1880 is
the omitted year, and the vertical bars represent the 95% confidence intervals.
manufacturing production could become more concentrated in places that obtained

national banks, leading to reallocation of manufacturing production away from the
surrounding places.
We estimate the same dynamic differences-in-differences specification as in Equation
2.4 above, but replace the outcome variable to the changes in manufacturing production
per capita of adjacent counties. We find positive point estimates for neighboring counties’
manufacturing production per capita in 1890 and 1900, but the effects are noisy (see
Figure 2.9), suggesting some positive net effects of national banks on adjacent counties.
112
Figure 2.9: Spillover effect on neighboring counties
1000
manufacturing prod per capita
0 -500500
1860 1870 1880 1890 1900
Reduced form IV
N = 1899
Notes: Figure 2.9 shows the dynamic diff-in-diff coefficients for the reduced form and the IV estimates of
the effect of having a national bank on neighboring counties’ manufacturing production value. The
specification for the IV estimates is
Yist = Âk b k e (National Bank) ⇥ {year=k} + Âk gk X 0 ⇥ {year=k} + hst + ai + # it . 1880 is the omitted year,
and the vertical bars represent the 95% confidence intervals.
113
2.5 Conclusion
Financial intermediaries can facilitate economic growth through both the asset-side
channel, such as providing long-term investment credit and short-term working capital,
and the liability-side channel, such as issuing secure bank liabilities that could facilitate
transactions across space and over time. This paper studies the late 19th century United
States after the passage of the National Bank Act of 1864, and exploits a population-based
capital requirement of national banks to study the causal effects of bank entry on the
local economy, as well as provide empirical evidence on the importance of stable bank
liabilities in the real economy.
We establish that a lower regulatory capital requirement defined discretely by town
population strongly and robustly predicts a higher likelihood of national bank entry. The
national banks had little impact on overall agricultural production, access to long-term
credit (proxied by value of farmland and fixtures), or short-term credit (proxied by
fertilizers expenditure). However, production significantly shifted from non-traded crops
to traded crops, which provides evidence that the stability of national bank liabilities had
first-order effects on the local economic development.
More generally, we find that local trade activity increased following national bank
entry, which may especially benefit the manufacturing sector growth as manufacturing
outputs are traded goods. We find significant and persistent greater increase in man-
ufacturing production per capita following national bank entry. While manufacturing
inputs, and employment also increased, capital did not significantly scale up. The results
indicate that national banks extended limited long-term credit for capital acquisition, and
the effect of national bank entry on manufacturing sector growth comes from lowering
transactions costs with stable national bank notes and short-term credit provision. The
increased trade activity facilitated by national bank’s secure liabilities could also expose
manufacturings to greater market access and therefore encourage in-firm innovation. We
114
find that innovation activity, proxied by patents granted, also increased significantly after
gaining a national bank. Together, the initial significant growth in the manufacturing
sector as well as increased trade and innovation could contribute to the persistently
higher level of manufacturing production for at least two decades. In sum, our results
suggest that the newly entered national banks facilitated local economic development by
providing safe liabilities with stable value across places and over time.
115
3 | Safety in the Gold Standard Era
3.1 Introduction
The modern global financial system is characterized by having a central country that
provides both the dominant currency and a large fraction of the financial assets de-
nominated in that currency. Since Bretton Woods, the United States has filled that role,
and there is extensive empirical evidence documenting world demand for dollar assets
and in particular, safe dollar assets (Bernanke, 2005; Caballero and Krishnamurthy, 2009;
Maggiori, 2017). The US’s ability to borrow at lower rates because of this demand has
been deemed an “exorbitant privilege" since the Bretton Woods period (Gourinchas and
Rey, 2007; Jiang et al., 2019; Krishnamurthy and Vissing-Jorgensen, 2012; Maggiori, 2017).1
In this literature, the pre-WWI classical gold standard era has often been described as a
historical precedent to the current international monetary system, bearing all the same
hallmarks. In that sense, this system is understood to have persistent features that speak
to a global equilibria regardless of the particular identity of the central country. However,
to my knowledge, there has been no study showing that the empirical patterns found
in the post-Bretton Woods period actually hold historically, before the disruptions of
WWI. Thus understanding the similarities and differences between these systems would
be useful for understanding which institutional characteristics are truly relevant for the
functioning of the international monetary system.
1 The term was coined by Valéry Discard d’Estaing, the then French Minister of Finance, in the 1960s.
116
The assumption that the UK and US are analogues within their own historical periods
is reasonable given the broad institutional similarities between the current and pre-WWI
systems: historically, like the US, the UK had the most developed and deepest financial
system, the pound sterling was the primary currency used in both international trade and
sovereign and corporate debt issuance, and the decades before WWI were characterized
by low capital and goods flows restrictions. However, the gold standard system of fixed
exchange rates and low values of short-term government debt are also strikingly different
from the floating rates and significant levels of government debt found today. Indeed,
while the historical consensus is that the British, acting through the Bank of England,
directed the international gold standard, there is actually little agreement on how this
system even operated (Eichengreen and Flandreau, 1997).
In this paper, I take as the departure point the accepted idea that the United Kingdom,
and in particular the Bank of England, was the center of the international monetary
system. I then interpret the historical gold standard institutions in the context of modern
macrofinance and ask whether the empirical patterns found in Krishnamurthy and
Vissing-Jorgensen (2012) regarding the demand for safe assets holds. I focus on this
feature because it is the linchpin for predictions about the central country’s role as a
global insurer (Maggiori, 2017) and foreign exchange rate dynamics (Jiang et al., 2019).
I document a demand for safe assets where the supply of safety is captured by the
gold holdings at the central bank. This is the relevant measure because these holdings
constrain the central bank’s ability to discount short-term securities in London. If the
central bank were to discount more than this amount, it could only do so by printing
bank notes not backed in gold, which would break the gold standard. The price of safety
is captured by the spread between the short-term bills that were eligible to be discounted
at the Bank of England (“bank bills") and the short-term bills that were not (“trade bills").
Both are forms of private money, but the former has a central bank guarantee attached
to it. Using monthly yields collected from the Economist magazine from 1870–1914 for
117
securities with 3, 4, and 6 month maturities, I show that the supply of safety is strongly
negatively correlated with the price of safety (i.e. the spread). Since the Bank of England
did not discount any bills with 4 or 6 month maturity, the lack of a clear relationship in
that case acts as a placebo for the main findings.
The main finding is robust to the Nagel (2016) critique that the levels of the interest
rates are driving the spread. Controlling for the level of the non-safe asset yield or the
Bank of England’s discount rate do not affect the empirical pattern. This is in part because
of the gold standard adjustment process: the Bank of England raised its discount rate
when gold was low in order to attract gold inflows, which also raised interest rates in the
open market. High levels of interest rates are correlated with lower spreads, which is the
opposite concern.
In addition, the historical context provides an opportunity to use exogenous shocks to
gold mining output as an instrument for the supply of safety. I use deviations in rainfall
as an instrument for gold mining output. I focus on the time period from 1890 to 1914
because those are the years that South Africa dominates world gold production. The
gold mining industry in the Witwatersrand fields was unique in that output was almost
solely a function of labor. The Chamber of Mines was established to enforce a monopsony
over native labor, and laborers were heavily recruited. During droughts and agricultural
depressions, the native population was more willing to work in the mines because of the
diminished value of their outside option. The instrument can be applied both regionally
and for the entire nation.
The rest of the paper is organized as follows. Section 3.2 discusses the institutional
context of the classical gold stanard era. Section 3.3 explains the data collected. Section
3.4 presents the empirical results on the relationship between the supply of safety and
the relative price of safety. Section 3.5 concludes.
118
3.2 Institutional context
In this section, I discuss the institutional details of the pre-WWI gold standard in the
United Kingdom. I argue that gold was not only the monetary base but also the measure
of safety in the financial system. First, I discuss how the gold standard was implemented
in practice through rules that tied bank note issuance to the gold holdings at the Bank
of England. Second, I discuss how short-term commercial bills of exchange were the
primary liquid asset, but that their quality as a safe asset was determined by the gold
held as assets by the central bank.
3.2.1 The functioning of the gold standard
The role of gold as the monetary base is usually dated to beginning in 1821, when Britain
returned to specie payments at the pre-Napoleonic War parity (that had held since 1717).
The Bank Charter Act of 1844 gave the Bank of England the exclusive right to note
issuance, and it required that issuances above the fiduciary limit of had to be backed
1:1 in gold.2 Since the Bank of England was still a private institution and continued to
act as a bank to commercial interests in London, it took on its new responsibilities by
splitting into the Banking and Issue Departments. The Banking department continued
to engage in the normal business of banking, such as issuing deposits and discounting
bills of exchange, while the Issue department held the gold stock that backed the money
supply. By the 1860s, the vast majority of the money supply in England was in Bank of
England notes while only a small fraction was in coined gold.
A simplified version of the Banking Department’s balance sheet is show in Figure 3.1.
2 The monopoly on note issuance was only granted for England and Wales; Scotland and Ireland
continued to allows private banks to issue bank notes. Private banks that had already established a note
issuance were grandfathered in and allowed to deplete this supply naturally over time. The fiduciary
limit was initially set to £14 million, but it was raised to £17 million. £11 million is the amount that the
government owed the Bank of England from its original loan, and the remaining £6 million difference was
the amount that was authorized once the Bank took over the lapsed private bank notes. Therefore any
issuance above £17 million was backed 100% in gold.
119
The Asset side is comprised of the bank’s investments in government and other securities,
the securities that were used as collateral for discounts and advances, and what was
called the “reserve" of notes and coin.3 While 40–50% of these reserves were in gold and
the rest in notes, the total note supply was always above the fiduciary issue, so the notes
held by the Banking department were fully backed by gold in the Issue department. The
Liabilities side was comprised of deposits by the bank’s customers (including the UK and
other governments) and the capital of its shareholders.4 The Issue Department balance
sheet is simply gold coin and bullion (assets) and bank notes above the fiduciary issue
(liabilities). Bank notes held by the public were marked as being in circulation; the rest
was held by the Bank of England in the Banking Department.
Figure 3.1: Banking Department balance sheet
Assets Liabilities
Securities: government Deposits: government

Securities: other Deposits: banks
Discounts & advances Deposits: other
Reserve: notes & bullion Capital
Total Total
Notes: Figure 3.1 is a stylized balance sheet for the Bank of England’s Banking Department.
Under the Bank Charter Act of 1844, the domestic money supply was entirely governed
by the volume of gold held by the Issue Department at the Bank of England. Its appeal
then (and now to those distrustful of monetary authorities) was that keeping the money
supply out of the realm of policymakers would lead to price stability. Given a domestic
currency’s nominal peg to gold, if the economy grew faster than the supply of gold, the
3 In contrast to the modern terminology of reserves being a central bank liability, here the term essentially
refers to cash on hand, much like the reserves of a private bank.
4 The Bank of England was established as a private institution in 1694 with shareholder capital, and it
was not nationalized until 1946.
120
real price of gold would increase and incentivize mining, thereby increasing the supply
and stabilizing prices (Barro, 1979). However, the mining response can be quite slow, as
evidenced by the twenty years of deflation from 1873 to 1896.
3.2.2 Private money creation and the supply of safety
In this section, I argue that role of gold as a measure of safety in the financial system is
tied to the role of private money creation in the British financial market. The relevant
institutional context involves three factors: the established tradition of using bills of
exchange (analogous to short-term commercial paper) as a store of value and medium of
exchange, the relative scarcity of short-term government debt, and the Bank of England’s
implicit role as the Lender of Last Resort in the second half of the 19th century. Short-term
commercial bills were the main liquid assets in the financial sector, and ones known as
“bank bills" were eligible to be brought to the Bank of England’s Discount Window. This
central bank backstop essentially removed the credit risk on these instruments and made
them the safe assets of the day. However, the Bank of England’s constraint on meeting the
demand at its Discount Window was the amount of cash on hand it held as assets. Excess
demand could only be met by printing notes that were not fully backed by gold, marking
a departure from the gold standard. I discuss this safety constraint in more detail below.
First, the financial instrument known as a bill of exchange, which is essentially a
forward-dated IOU written by the payer and accepted by the payee, had been widely
used in Britain in commercial transactions for well over a century. In Lancashire, bills of
exchange almost completely replaced small denomination notes and coins as a substitute
for money in the earlier part of the 19th century (Ashton, 1945). Mechanically, anyone
who received a bill of exchange as payment for a good or service could use it to pay
for goods/services in turn by having his creditor accept it. The bill would continue to
circulate until it finally matured, usually one to three months after it was issued, and the
original debt cleared between the payer and the ultimate payee.
121
The private money creation allowed by bills of exchange was recognized by contem-
poraries as having the same sorts of consequences for financial stability as described by
Stein (2012). In fact, the 1772 British financial crisis was recognized as being caused by a
troubled bank, the Ayr Bank, writing bills of exchange purely to extend their own credit
lines (Kosmetatos, 2018). In addition, Henry Thornton pointed out that even if the bill
was written for a real underlying commercial transaction (what was known as the real
bills doctrine), the nominal value of the obligations it represented was multiplied by the
number of times the bill was passed on (Thornton, 1802). As a result, the Bank of England
adhered to the norm of only discounting bills that represented real economic activity and
had not been accepted more than twice.
Merchant banks with well established reputations had a natural advantage in creating
private money. The interest rate at which the payers borrowed from their creditors during
the period of the bill depended on his credit risk. Merchant banks that could borrow at
low rates therefore began to sell their guarantees for a fee to their own customers. The
bills that they accepted from their customers became known as “banker’s acceptances"
and were considered the safest short-term assets in financial markets. As the banker’s
acceptances became more common, they were increasingly used to finance international
trade around the world. British multinational banks operating in cities around the world
accepted their customers’ bills and remitted them back to London, where there was strong
demand from domestic banks. Banks held these instruments as a form of collateral or as
interest-bearing investments that could easily be converted into cash to meet depositor
demand.5
Second, until WWI, short-term government debt was a negligible component of debt,
and most of the UK’s debt was in the form of long-term consol bonds. Indeed, the idea of
the Treasury bill was first suggested by Walter Bagehot and modeled after the commercial
bill of exchange to allow the government to also borrow short-term at low rates. However,
5 See Xu (2018) for a lengthy discussion of the role of British multinational banks in providing interna-
tional trade finance.
122
the UK government barely used this instrument until WWI. Estimates of the volumes of
short-term commercial bills dwarf the supply of government debt (Nishimura, 1971b).
Third, the financial crises that occurred in 1847, 1857, and 1866 successively cemented
the Bank of England’s role as a Lender of Last Resort.6 While the Bank maintained strict
rules about which securities were eligible for the Discount Window, its willingness to
discount those securities meant that they were the safest asset in the economy. However,
the securities that the Bank could discount was limited by the size of its balance sheet.
The Banking Department’s balance sheet could not expand without issuing more deposits
or raising more capital, so increasing its Discounts necessarily required either reducing
its holdings of Securities (primarily government) or else reducing its Reserves of cash
and gold.
Figure 3.2 plots the time series of the primary components of the asset side of the
Banking department’s balance sheet. This series runs from 1870 to 1914, the decades
that are considered to be the classical gold standard era. Only in 1890 does it appear
that the increase in Discounts was funded by selling Securities. For most of the period,
and particularly after 1895, there is a strong negative correlation between the amount of
lending that the Bank engaged in at its Discount window and its Reserve holdings. Given
both the institutional context and the empirical pattern, I therefore argue that the gold
reserves at the Bank of England as the constraint on its ability to guarantee the safety of
this form of private money.
3.2.3 South Africa mining
The majority of London’s gold came from its overseas colonies, such as Canada, Australia,
New Zealand, and South Africa. In South Africa, the Witwatersrand gold fields near
Johannesburg are the largest in the world, and by 1914 accounted for almost 50% of
6 During the crises listed, the demand for Discounts was so high that the Bank of England obtained
permission from the Chancellor of the Exchequer to suspend the Bank Charter Act and thereby suspend
the gold standard. Although the Bank obtained this permission, they did not actually use it.
123
Figure 3.2: Composition of Assets in the Banking Department (1870–1914)
80
60
millions GBP
40
20
0
1870 1875 1880 1885 1890 1895 1900 1905 1910 1915
Year
Securities Reserve (cash on hand)

Discount window Total Assets
Total Assets = Securities + Discount Window + Reserve
Notes: Figure 3.2 plots the main categories of Assets in the Bank of England’s Banking Department from
1870 – 1914. Source: Bank of England Archives C/1 Daily Accounts ledgers.
124
annual world production and 80% of British gold imports. Their discovery and subsequent
development in the 1890s provided a significant, steady source of gold to London for
twenty years. Unlike the fields in Canada and Australia that could be panned by hand,
South African gold was very finely distributed in rock and could only be extracted by a
process of drilling, crushing, and chemically treating the ore to release the gold content.
During the early period of development, there was much uncertainty about the nature
and size of the gold fields and whether the fields would be profitable. However, as the
mines were developed, it became clear that labor was the binding constraint because vast
volumes of ore had to be processed for small amounts of gold (Jeeves, 1985).
The main source of labor was the native South African population, although this
population also had the outside option of living in their native areas. Conditions in the
minefields were poor, with long hours, low pay, and little healthcare. The Chamber of
Mines was an organization comprised of the individual mining companies that set wages
and conditions such that they created a monopsony over labor. Their main difficulty
was balancing their need to attract and recruit labor with their desire to keep wages low.
According to the annual reports of the Chamber of Mines, native labor was almost in
constant shortage, to the extent that they resorted to important migrant labor from China
for many years (Richardson, 1982).
Despite the Chambers’ policies for importing and coercing labor to work on the
minefields, production was almost continually constrained by the labor supply the entire
period.7 I propose to use the annual deviations from ideal rainfall levels as an instrument
for gold mining production, operating through the value of the agricultural (outside)
sector available to native laborers. Since South Africa had no other industries besides
mining and agriculture (which was used for domestic consumption rather than exported),
it is unlikely that the rainfall affected the economy through other channels. While
the main specification focuses on the annual deviations in South Africa, I can provide
7 This relationship is borne out both in the contemporary reports on the industry as well as in the data
plotting the correlation between labor supply and mining output.
125
robustness from subregions within the country. In addition, other first stage outcomes
include native labor recruited and total labor on the mines.
3.3 Data
In this section, I describe both the data that I use for the main empirical exercise as well
as the desired data for testing predictions generated by the modern safe assets literature.
3.3.1 Finance data
The financial markets data come from the Bank of England’s archives and historical issues
of the Economist magazine. The Bank of England published daily, weekly, and monthly
summaries of its balance sheet. The disaggregated balance sheet is kept in ledgers at the
Bank of England’s archives.
I collected the details of the weekly (Friday) composition of assets and liabilities in the
Bank of England’s Banking and Issue departments from 1870–1914. I also collected the
yields on short-term commercial paper printed daily in the Economist magazine during
this period for the same dates. These included rates on 3, 4, and 6 month bank bills and
trade bills. The Bank of England’s official discount rate, known as the Bank Rate, was
also collected for these dates. Figure 3.3 panels A and B show the Bank’s balance sheet
for the week of March 24, 1882 and the rates published in the Economist for that week.
3.3.2 South African data
The data on South African mining come from a variety of sources. Government statistics
provide aggregate mining output and exports to the UK annually. Weather stations
in South Africa also recorded monthly rainfall in different areas of the country. The
Chamber of Mines produced detail annual reports with monthly labor and production
figures for each mine.
126
Figure 3.3: Data sources
(a) Bank of England balance sheet
(b) Economist reports on rates
Notes: Figure 3.3 A shows a page from The Bank of England Archive C/1 Ledger for February 1882. Figure
3.3 B shows the yields on 3, 4, and 6 month maturity Bank Bills and Trade Bills published weekly in the
Economist.
127
3.4 Demand for safe assets
3.4.1 Main results using OLS
The empirical strategy follows Krishnamurthy and Vissing-Jorgensen (2012), where I first
estimate the elasticity of the spread to the supply of gold reserves:
ittrade itbank = a + b 0 Goldt + GXt + # t (3.1)
The variables ittrade and itbank refer to the yield on trade bills and bank bills, respectively.
Goldt is the value of gold at the Bank of England, normalized by GDP.8 Xt denotes a
vector of control variables which includes seasonality dummies as well as other controls
such as the Bank of England discount rate.
I estimate this regression on monthly data (using the end-of-month dates) from 1870
to 1914 and plot the relationship in Figure 3.4. These binscatters show a clear negative
relationship between the supply of central bank gold and the spread between the assets
eligible to be discounted and those that are not. In the first row, I focus on the spread
for the three month maturity securities. The first figure controls only for seasonal flows
while the second figure also controls for the policy discount rate.
In the second row, I plot the relationship for four and six month securities, respectively.
The relationship remains negative at 4 months, but becomes positive at 6 months. These
relationships are robust to a number of different normalizations and variables used to
calculate the supply of safety (for instance, the supply of gold only in the reserves and
total reserves including bank notes). Table 3.1 reports the coefficients and standard errors
from the linear relationship using Newey-West standard errors with 12 lags.
The main coefficients of interest are in columns 1 and 2, which report the relationship
8 The demand for safety is assumed to be proportional to the size of the domestic economy. In the
baseline specification, I use the total (nominal) value of gold at the central bank.
128
Figure 3.4: Relationship between supply of Gold/GDP and safety premium
Trade bill - Bank bill spread

3 month maturity 3 month maturity
.6 .6
.5 .5
.4 .4
.3 .3
.15 .2 .25 .3 .15 .2 .25 .3

Gold reserves/GDP Gold reserves/GDP
residualized on month dummies residualized on month dummies and Bank Rate

.6 .6
.55 .55
.5 .5
.45 .45
.4 .4
.15 .2 .25 .3 .15 .2 .25 .3

.7 .75
.65 .7
.65
.6
.6
.55 .55
.5 .5
.15 .2 .25 .3 .15 .2 .25 .3
Notes: Figure 3.4 plots the binscatter and line of best fit from the OLS regression using monthly data from
1870–1914. The dependent variable is the spread between the yields on the safest asset (“Bank" bills) and
the second safest asset (“Trade" bills). Gold is the total volume of bullion at the Bank of England,
normalized by GDP. The Bank Rate is the Bank of England’s reported discount rate, which affected the total
level of rates in London. All specifications are residualized on monthly indicators to control for seasonality.
129
between the supply of safety and the price of safety for the assets with the relevant
maturity (discountable at the Bank of England): 3 months. The coefficients can be
interpreted as meaning a 1 pp increase in gold supply relative to GDP reduces the yield
spread by 1.1 basis points. This negative relationship is not purely driven by the level of
the interest rate, as shown in column 2 when we control for the Bank Rate.
Table 3.1: Relationship between supply of safety (gold) and safety spread at different maturities
3 month 4 month 6 month

(1) (2) (3) (4) (5) (6)
Gold -1.124** -1.560** -0.483 -0.497 0.606 1.225**
(0.555) (0.622) (0.411) (0.455) (0.512) (0.524)
Bank Rate -0.0348** -0.00113 0.0491***
(0.0150) (0.0165) (0.0176)
Month FE Y Y Y Y Y Y
N 524 524 524 524 534 534
Notes: Table 3.1 reports the estimates from the OLS regression using monthly data from 1870–1914. The
dependent variable is the spread between the yields on the safest asset (“Bank" bills) and the second safest
asset (“Trade" bills). Gold is the total volume of bullion at the Bank of England, normalized by GDP. The
Bank Rate is the Bank of England’s reported discount rate, which affected the total level of rates in London.
Columns 1 and 2 report estimates using securities with 3 month maturity; Columns 3 and 4 use 4 month
maturities, and Columns 5 and 6 use 6 month maturities. All specifications are residualized on monthly
indicators to control for seasonality. Standard errors are Newey-West with 12 lags. *p < 0.1, **p < 0.05,
***p < 0.01
3.5 Conclusion
The historical precedent of the British-led gold standard provides a valuable opportunity
to gain additional insight into the way the International Monetary System functions. A
demand for US dollar-denominated safe assets is a striking pattern in today’s economy,
and it appears that there is a historical analogue in the demand for British gold-backed
short-term securities. In future work, it will be possible to use plausibly exogenous shocks
to gold mining output to isolate this demand in a way that is not possible in today’s world.
130
In addition, macrofinance models on the role of the central country provide additional
predictions that can be further tested historically, lending a broader and more externally
valid understanding of the International Monetary System.
131
References
Abadie, A., Athey, S., Imbens, G. W. and Wooldridge, J. (2017). When should you adjust
standard errors for clustering? Tech. rep., NBER Working Paper.
Adão, R., Kolesár, M. and Morales, E. (2018). Shift-share designs: Theory and inference.
Tech. rep., NBER Working Paper.
Admati, A. and Hellwig, M. (2014). The Bankers’ New Clothes: What’s Wrong with Banking
and What to Do about It. Princeton University Press.
Ahn, J., Amiti, M. and Weinstein, D. E. (2011). Trade finance and the great trade collapse.
American Economic Review, 101 (3), 298–302.
Ales, L., Carapella, F., Maziero, P. and Weber, W. E. (2008). A model of banknote
discounts. Journal of Economic Theory, 5 (27).
Alessandria, G., Kaboski, J. P. and Midrigan, V. (2010). The great trade collapse of
2008–09: An inventory adjustment? IMF Economic Review, 58 (2), 254–294.
Allen, T. and Donaldson, D. (2018). Geography and path dependence. Tech. rep.
Amiti, M. and Weinstein, D. E. (2011). Exports and financial shocks. Quarterly Journal of
Economics, 126 (4), 1841–1877.
— and — (2018). How much do idiosyncratic bank shocks affect investment? evidence
from matched bank-firm loan data. Journal of Political Economy, 126 (2), 525–587.
Anson, M., Bholat, D., Kang, M., Rieder, K. and Thomas, R. (2018). Frosted glass or
raised eyebrow? Central bank credit rationing and the Bank of England’s Discount Window
policies during the Crisis of 1847. Tech. rep., Working paper.
Antràs, P. and Foley, C. F. (2015). Poultry in motion: A study of international trade

finance practices. Journal of Political Economy, 123 (4), 853–901.
Ashcraft, A. B. (2005). Are banks really special? new evidence from the fdic-induced
failure of healthy banks. American Economic Review, 95 (5), 1712–1730.
Ashton, T. S. (1945). The bill of exchange and private banks in lancashire, 1790-1830. The
Economic History Review, 15 (1/2), 25–35.
132
Atack, J. (2016). Historical geographic information systems (gis) database of u.s. railroads
for 1870 to 1890.
Auboin, M. (2012). Financing international trade. In H. K. Baker and L. A. Riddick (eds.),
International Trade Finance: A Survey, 11, Oxford University Press, pp. 231–247.
Bagehot, W. (1873). Lombard Street: a description of the money market. Scribner, Armstrong
& Company.
Baldwin, R. (1988). Hysteresis in import prices: The beachhead effect. American Economic
Review, 78 (4), 773–783.
— and Krugman, P. (1989). Persistent trade effects of large exchange rate shocks. Quarterly
Journal of Economics, 104 (4), 635–654.
Barro, R. J. (1979). Money and the price level under the gold standard. The Economic
Journal, 89 (353), 13–33.
Baster, A. (1934). The origins of british banking expansion in the near east. The Economic
History Review, 5 (1), 76–86.
Beckert, S. (2015). Empire of cotton: A global history. Vintage.
Benmelech, E., Frydman, C. and Papanikolaou, D. (2016). Credit market disruptions and
employment during the Great Depression: Evidence from firm-level data. Tech. rep., Working
paper.
Berman, N. and Héricourt, J. (2010). Financial factors and the margins of trade: Evidence
from cross-country firm-level data. Journal of Development Economics, 93 (2), 206–217.
Bernanke, B. S. (1983). Irreversibility, uncertainty, and cyclical investment. Quarterly
Journal of Economics, 98 (1), 85–106.
— (2005). The global saving glut and the US current account deficit. Tech. rep.
— and Gertler, M. (1989). Agency costs, net work, and business fluctuations. American
Economic Review, 79 (1), 14–31.
Blattner, L., Farinha, L. and Rebelo, F. (2018). When losses turn into loans: The cost of
undercapitalized banks. Tech. rep., Working paper.
Bleakley, H. and Lin, J. (2012). Portage and path dependence. Quarterly Journal of
Economics, 127 (2), 587–644.
BOE (1964). The Treasury Bill. Tech. Rep. 3, Bank England Quarterly Bulletin.
Bordo, M. D. (1998). Currency crises (and banking crises) in historical persepctive.
Stockholm School of Economics Research Report, (10).
— and Haubrich, J. G. (2010). Credit crises, money and contractions: An historical view.
Journal of Monetary Economics, 57 (1), 1–18.
133
Borusyak, K., Hull, P. and Jaravel, X. (2018). Quasi-experimental shift-share designs. Tech.
rep., NBER working paper.
Bougheas, S. (2004). Internal vs external financing of r&d. Small Business Economics, 22 (1),
11–17.
Briones, I. and Rockoff, H. (2005). Do economists reach a conclusion on free-banking

episodes? Economic Journal Watch, 2 (2).
Bruland, K. and Mowery, D. C. (2006). Innovation through time. The Oxford Handbook of
Innovation.
Caballero, R. J. and Krishnamurthy, A. (2009). Global imbalances and financial fragility.

Cagan, P. (1963). The first fifty years of the national banking system. Banking and Monetary
Studies, edited by Deane Carson, 15, 42.
Calomiris, C. W. and Carlson, M. (2018). Bank examiners’ information and expertise

and their role in monitoring and disciplining banks before and during the panic of
1893.
— and Mason, J. R. (2003). Consequences of bank distress during the great depression.
Cameron, A. C. and Miller, D. L. (2014). Robust inference for dyadic data. Tech. rep.,
Working paper.
Carlson, M., Correia, S. and Luck, S. (2018). The effects of banking competition on
growth and financial stability: Evidence from the national banking era.
Carter, S. B. and Sutch, R. (1996). Fixing the facts: Editing of the 1880 us census of
occupations with implications for long-term labor-force trends and sociology of official
statistics. Historical Methods, 29 (1), 5–24.
Cassis, Y. (2016). Private banks and private banking. The Oxford Handbook of Banking and
Financial History, pp. 87–109.
Cerra, V. and Saxena, S. C. (2008). Growth dynamics: The myth of economic recovery.
Cetorelli, N. and Goldberg, L. S. (2011). Global banks and international shock trans-
mission: Evidence from the crisis. IMF Economic Review, 59 (1), 41–76.
Chaney, T. (2016). Liquidity constrained exporters. Journal of Economic Dynamics and

Control, 72, 141–154.
Chapman, S. D. (1984). The rise of merchant banking. Taylor & Francis.
134
Chodorow-Reich, G. (2014). The employment effects of credit market disruptions: Firm-
level evidence from the 2008–9 financial crisis. Quarterly Journal of Economics, 129 (1),
1–59.
Chor, D. and Manova, K. (2012). Off the cliff and back? credit conditions and interna-
tional trade during the global financial crisis. Journal of International Economics, 87 (1),
117–133.
Davis, D. R. and Weinstein, D. E. (2002). Bones, bombs, and break points: The geography
of economic activity. American Economic Review, 92 (5), 1269–1289.
Dedinger, B. and Girard, P. (2017). Exploring trade globalization in the long run: The
ricardo project. Historical Methods: A Journal of Quantitative and Interdisciplinary History,
50 (1), 30–48.
Dewey, D. R. (1910). State banking before the civil war. National Monetary Commission, 61st
Congress, 2nd Session.
Diamond, D. W. and Dybvig, P. H. (1983). Bank runs, deposit insurance, and liquidity.
Journal of political economy, 91 (3), 401–419.
Donaldson, D. and Hornbeck, R. (2016). Railroads and american economic growth: A

“market access” approach. The Quarterly Journal of Economics, 131 (2), 799–858.
Eaton, J., Kortum, S., Neiman, B. and Romalis, J. (2016). Trade and the global recession.
Eichengreen, B. and Flandreau, M. (1997). The gold standard in theory and history.
Psychology Press.
— and Rose, A. K. (2004). Staying afloat when the wind shifts: External factors and
emerging-market banking crises. Money, Capital Mobility, and Trade: Essays in Honor of
Robert A. Mundell, p. 171.
Einzig, P. (1931). The fight for financial supremacy. London: Macmillan and Company.
Feenstra, R. C., Li, Z. and Yu, M. (2014). Exports and credit constraints under incomplete
information: Theory and evidence from china. Review of Economics and Statistics, 96 (4),
729–744.
Feigenbaum, J. J. and Rotemberg, M. (2014). Information and investment: Impacts of the

introduction of rural free delivery.
Flandreau, M. and Ugolini, S. (2013). Where it all began: lending of last resort and the Bank
of England during the Overend, Gurney Panic of 1866. Tech. rep., Norges Bank working
paper.
Fouquin, M. and Hugot, J. (2016). Two centuries of bilateral trade and gravity data: 1827-2014.
Tech. rep., Universidad Javeriana-Bogotá.
135
Friedman, M. and Schwartz, A. J. (1965). A monetary history of the United States, 1867-1960.
Princeton University Press.
Frydman, C., Hilt, E. and Zhou, L. Y. (2015). Economic effects of runs on early “shadow
banks”: Trust companies and the impact of the panic of 1907. Journal of Political Economy,
123 (4), 902–940.
Fulford, S. L. (2015). How important are banks for development? national banks in the
united states, 1870–1900. Review of Economics and Statistics, 97 (5), 921–938.
Gan, J. (2007). The real effects of asset market bubbles: Loan-and firm-level evidence of a
lending channel. The Review of Financial Studies, 20 (6), 1941–1973.
Goldsmith-Pinkham, P., Sorkin, I. and Swift, H. (2018). Bartik instruments: What, when,
why, and how. Tech. rep., NBER working paper.
Gorton, G. (1996). Reputation formation in early bank debt markets. Journal of Political
Economy, 104 (2), 346–397.
— (1999). Pricing free bank notes. Journal of Monetary Economics, 44, 33–64.
— and Pennacchi, G. (1990). Financial intermediaries and liquidity creation. The Journal
of Finance, 45 (1), 49–71.
Gou, M. (2016). Did capital requirements promote bank stability in the early 20th century
united states?
Gourinchas, P.-O. and Rey, H. (2007). International financial adjustment. Journal of

Political Economy, 115 (4), 665–703.
—, — and Truempler, K. (2012). The financial crisis and the geography of wealth transfers.
Journal of International Economics, 88 (2), 266–283.
Grada, C. O. and White, E. N. (2003). The panics of 1854 and 1857: A view from the
emigrant industrial savings bank. The Journal of Economic History, 63 (1), 213–240.
Greenstone, M., Mas, A. and Nguyen, H.-L. (2014). Do credit market shocks affect the real
economy? Quasi-experimental evidence from the Great Recession and ‘normal’ economic times.
Tech. rep., NBER working paper.
Hall, B. H. (2002). The financing of research and development. Oxford Review of Economic
Policy, 18 (1).
Head, K. and Mayer, T. (2014). Gravity equations: Workhorse, toolkit, and cookbook. In
Handbook of International Economics, vol. 4, Elsevier, pp. 131–195.
Helpman, E., Melitz, M. and Rubinstein, Y. (2008). Estimating trade flows: Trading
partners and trading volumes. The quarterly journal of economics, 123 (2), 441–487.
136
Hornbeck, R. (2010). Barbed wire: Property rights and agricultural development. Quar-
terly Journal of Economics, 125 (2), 767–810.
Huber, K. (2018). Disentangling the effects of a banking crisis: Evidence from german
firms and counties. American Economic Review, 108 (3), 868–98.
Iacovone, L. and Zavacka, V. (2009). Banking crises and exports: Lessons from the past
for the recent trade collapse. The Great Trade Collapse: Causes, Consequences and Prospects,
p. 107.
Iyer, R., Peydró, J.-L., da Rocha-Lopes, S. and Schoar, A. (2013). Interbank liquidity
crunch and the firm credit crunch: Evidence from the 2007–2009 crisis. The Review of
Financial Studies, 27 (1), 347–372.
Jaremski, M. (2014). National banking’s role in u.s. industrialization, 1850-1900. Journal of

Economic History, 74, 109–140.
Jayaratne, J. and Strahan, P. E. (1996). The finance-growth nexus: Evidence from bank
branch deregulation. Quarterly Journal of Economics.
Jeeves, A. H. (1985). Migrant labour in South Africa’s mining economy: the struggle for the
gold mines’ labour supply, 1890-1920. McGill-Queen’s Press-MQUP.
Jenks, L. H. (1927). The migration of British capital to 1875. AA Knopf.
Jiang, Z., Krishnamurthy, A. and Lustig, H. N. (2019). Dollar safety and the global
financial cycle. Available at SSRN 3328808.
Jones, G. (1995). British multinational banking, 1830-1990. Oxford University Press.
— (2000). Merchants to multinationals: British trading companies in the nineteenth and twentieth
centuries. Oxford University Press.
Juhász, R. (2018). Temporary protection and technology adoption: Evidence from the
napoleonic blockade. American Economic Review, 108 (11), 3339–76.
Juhász, R. and Steinwender, C. (2017). Drivers of fragmented production chains: Evidence

from the 19th century. Tech. rep., working paper.
Kaminsky, G. L. and Reinhart, C. M. (1999). The twin crises: The causes of banking and
balance-of-payments problems. American economic review, 89 (3), 473–500.
Kashyap, A. K., Rajan, R. G. and Stein, J. C. (2002). Banks as liquidity providers: An

explanation for the coexistence of lending and deposit-taking. The Journal of Finance,
LVII (1), 33–73.
Khwaja, A. I. and Mian, A. (2008). Tracing the impact of bank liquidity shocks: Evidence
from an emerging market. American Economic Review, 98 (4), 1413–1442.
King, W. T. (1936). History of the London discount market. Routledge.
137
Kisling, W. A. (2017). Trade finance in historical perspective: The role of German banks in the
rise of Germany as international trade power, 1875-1913. Tech. rep., Working paper.
Kiyotaki, N. and Moore, J. (1997). Credit cycles. Journal of Political Economy, 105 (2),
211–248.
Kline, P. and Moretti, E. (2014a). Local economic development, agglomeration

economies, and the big push: 100 years of evidence from the tennessee valley au-
thority. Quarterly Journal of Economics, 129 (1), 275–331.
— and — (2014b). Local economic development, agglomeration economies, and the big
push: 100 years of evidence from the tennessee valley authority. The Quarterly Journal of
Economics, 129 (1).
Knox, J. J. (1900). A History of Banking in the United States. BradFord Rhodes & Company.
Kosmetatos, P. (2018). The 1772–73 British Credit Crisis. Springer.
Krishnamurthy, A. and Muir, T. (2017). How credit cycles across a financial crisis. Tech.
rep., NBER working paper.
— and Vissing-Jorgensen, A. (2012). The aggregate demand for treasury debt. Journal of
Political Economy, 120 (2), 233–267.
Landon-Lane, J. and Rockoff, H. (2007). The origin and diffusion of shocks to regional
interest rates in the united states, 1800-2002. Explorations in Economic History, 44 (3),
487–500.
Mackenzie, C. (2013). Realms of silver: one hundred years of banking in the East. Routledge.
Maggiori, M. (2017). Financial intermediation, international risk sharing, and reserve

currencies. American Economic Review, 107 (10), 3038–71.
Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate

industry productivity. Econometrica, 71 (6), 1695–1725.
Mian, A. (2006). Distance constraints: The limits of foreign lending in poor economies.
The Journal of Finance, 61 (3), 1465–1505.
Michie, R. C. (2016). British banking: continuity and change from 1694 to the present. Oxford
University Press.
Mitchener, K. J. and Weidenmier, M. (2008). Trade and empire. The Economic Journal,
118 (533), 1805–1834.
Muirhead, S. and Green, E. (2016). Crisis banking in the East: The history of the Chartered
Mercantile Bank of London, India and China, 1853–93. Routledge.
Nagel, S. (2016). The liquidity premium of near-money assets. The Quarterly Journal of
Economics, 131 (4), 1927–1971.
138
Nishimura, S. (1971a). The decline of inland bills of exchange in the London money market
1855-1913. Cambridge University Press.
— (1971b). The decline of inland bills of exchange in the London money market 1855-1913.
Cambridge University Press.
—, Suzuki, T., Michie, R. and Michie, R. C. (2012). The origins of international banking in
Asia: The nineteenth and twentieth centuries. Oxford University Press.
O’Rourke, K. H. and Williamson, J. G. (1999). Globalization and history. MIT Press.
Oster, E. (2017). Unobservable selection and coefficient stability: Theory and evidence.
Journal of Business & Economic Statistics, 0 (0), 1–18.
Paravisini, D., Rappoport, V., Schnabl, P. and Wolfenzon, D. (2014). Dissecting the
effect of credit supply on trade: Evidence from matched credit-export data. Review of
Economic Studies, 82 (1), 333–359.
Pascali, L. (2017). The wind of change: Maritime technology, trade and economic
development. American Economic Review, 107 (9), 2821–2854.
Peek, J. and Rosengren, E. S. (1997). The international transmission of financial shocks:

The case of japan. American Economic Review, 87 (4), 495–505.
— and — (2000). Collateral damage: Effects of the japanese bank crisis on real activity in
the united states. American Economic Review, 90 (1), 30–45.
Pennacchi, G. (2006). Deposit insurance, bank regulation, and financial system risks.
Journal of Monetary Economics, 53 (1), 1–30.
Petersen, M. A. and Rajan, R. G. (1994). The benefits of lending relationships: Evidence

from small business data. The Journal of Finance, 49 (1), 3–37.
— and — (2002). Does distance still matter? the information revolution in small business
lending. The Journal of Finance, 57 (6), 2533–2570.
Petralia, S., Balland, P.-A. and Rigby, D. (2016). Unveiling the geography of historical
patents in the united states from 1836 to 1975. Scientific Data.
Pope, J. E. (1914). Agricultural credit in the united states. The Quarterly Journal of Economics,
28 (4), 701–746.
Puri, M., Rocholl, J. and Steffen, S. (2011). Global retail lending in the aftermath of
the us financial crisis: Distinguishing between supply and demand effects. Journal of
Financial Economics, 100 (3), 556–578.
Quadrini, V. (2017). Bank liabilities channel. Journal of Monetary Economics, 89(C), 25–44.
Rajan, R. G. (1992). Insiders and outsiders: The choice between informed and arm’s-
length debt. The Journal of finance, 47 (4), 1367–1400.
139
— and Zingales, L. (1998a). Financial dependence and growth. American Economic Review,
88 (3), 559–586.
— and — (1998b). Financial dependence and growth. The American Economic Review, 88,
559–586.
Reber, V. B. (1979). British mercantile houses in Buenos Aires, 1810-1880. Harvard University
Press.
Redding, S. J. and Sturm, D. M. (2008). The costs of remoteness: Evidence from german
division and reunification. American Economic Review, 98 (5), 1766–97.
Reinhart, C. M. and Rogoff, K. S. (2009a). The aftermath of financial crises. American

Economic Review, 99 (2), 466–72.
— and — (2009b). This time is different: eight centuries of financial folly. Princeton University
Press.
Rey, H. (2015). Dilemma not trilemma: The global financial cycle and monetary policy indepen-
dence. Tech. rep., National Bureau of Economic Research.
Richardson, G. and Troost, W. (2009). Monetary intervention mitigated banking panics

during the great depression: Quasi-experimental evidence from a federal reserve district
border, 1929–1933. Journal of Political Economy, 117 (6), 1031–1073.
Richardson, P. (1982). Chinese mine labour in the Transvaal. Springer.
Rockoff, H. (1975). The Free Banking Era: A Reconsideration. New York: Arno Press.
— (1991). Unregulated Banking: Chaos or Order?, London: Macmillan Academic and

Professional Ltd., chap. Lessons from the American Experience with Free Banking.
Rolnick, A. J. and Weber, W. E. (1982). Free banking, wildcat banking, and shinplasters.
Federal Reserve Bank of Minneapolis Quarterly Review.
Romer, C. D. and Romer, D. H. (1989). Does monetary policy matter? a new test in the
spirit of friedman and schwartz. NBER macroeconomics annual, 4, 121–170.
— and — (2017). New evidence on the impact of financial crises in advanced countries.
— and — (2018). Why some times are different: Macroeconomic policy and the aftermath
of financial crises. Economica, 85, 1–40.
Rousseau, P. L. and Sylla, R. (2005). Emerging financial markets and early us growth.
Explorations in Economic History, 42 (1), 1–26.
— and Wachtel, P. (1998). Financial intermediation and economic performance: Histor-

ical evidence from five industrialized countries. Journal of Money, Credit and Banking,
30 (4), 657–678.
140
Sarkees, M. R. and Wayman, F. (2010). Resort to war: 1816-2007. Correlates of War. Wash-
ington DC: CQ Press.
Schnabl, P. (2012). The international transmission of bank liquidity shocks: Evidence
from an emerging market. The Journal of Finance, 67 (3), 897–932.
Schularick, M. and Taylor, A. M. (2012). Credit booms gone bust: Monetary policy,
leverage cycles, and financial crises, 1870-2008. American Economic Review, 102 (2),
1029–61.
Schwartz, A. J. (1987). Real and pseudo-financial crises. In Money in historical perspective,
University of Chicago Press, pp. 271–288.
Sequeira, S., Nunn, N. and Qian, N. (2018). Immigrants and the making of america.
Review of Economic Studies, Forthcoming.
Sharpe, S. A. (1990). Asymmetric information, bank lending, and implicit contracts: A
stylized model of customer relationships. The Journal of Finance, 45 (4), 1069–1087.
Sprague, M. (2011). Bodie’s Gold. University of Nevada Press.
Stein, J. C. (2012). Monetary policy as financial stability regulation. The Quarterly Journal
of Economics, 127 (1), 57–95.
Stephen K. Williams, L. (ed.) (1901). Reports of Cases Argued and Decided in the Supreme
Court of the United States (1843-1846). 11, The Lawyer’s Co-Operative Publishing Com-
pany.
Sylla, R. (1969). Federal policy, banking market structure, and capital mobilization in the
united states, 1863–1913. The Journal of Economic History, 29 (4), 657–686.
— (1982). Monetary innovation in america. The Journal of Economic History, 42 (1), 21–30.
Thornton, H. (1802). An enquiry into the nature and effects of the paper credit of Great Britain.
J. Hatchard.
Weber, W. E. (2003). Interbank payments relationships in the antebellum united states:
Evidence from pennsylvania. Journal of Monetary Economics, 50 (2), 455–474.
— (2012). Clearing arrangements in the united states before the federal reserve system.
Weiss, C. (2018). Contractionary devaluation risk: Evidence from u.s. silver coinage
agitation, 1878-1900.
White, E. N. (1982). The membership problem of the national banking system. Explorations
in Economic History, 19 (2), 110–127.
— (1998). Were banks special intermediaries in late nineteenth century america. Federal
Reserve Bank of St. Louis Review.
Xu, C. (2018). Reshaping global trade: the immediate and long-run effects of bank failures.
working paper.
141
A | Appendix to Chapter 1
A.1 Additional tables
Table A1: Bank-level relationship between failure and credit supply
D% Creditb
(1) (2) (3) (4)
Failureb -0.782*** -0.869*** -0.971*** -0.946***
[0.109] [0.126] [0.166] [0.157]
Weighting none Capital, 1865 Liabilities, 1865 Size, 1865
N 31 31 31 31
Adj. R2 0.398 0.413 0.511 0.488
Notes: Table A1 shows the regression results for the pseudo first stage relationship between bank failure
and the credit supplied. The dependent variable is the percent change in the trade credit supply of
individual banks. Banks that failed are given a trade credit supply of 0 in the post-crisis period. There are
31 banks that report the composition of their balance sheet. Column 1 reports the baseline, unweighted
regression. In columns 2-4, the regressions are weighted by different proxies for firm size. Robust standard
errors in brackets. *p < 0.1, **p < 0.05, ***p < 0.01
142
Table A2: Bank balance on geographic exposure calculated as percent of assets
UK % 0.09 (0.21) 0.10 (0.22) 0.06 (0.17) 0.04 (0.0)

Brit. Emp. % 0.39 (0.41) 0.42 (0.41) 0.28 (0.38) 0.14 (0.1)
Europe % 0.32 (0.39) 0.30 (0.38) 0.44 (0.39) -0.15 (0.1)
Asia % 0.26 (0.34) 0.25 (0.34) 0.26 (0.35) -0.01 (0.1)
Africa % 0.09 (0.21) 0.09 (0.21) 0.10 (0.21) -0.01 (0.0)
N. America % 0.16 (0.31) 0.18 (0.33) 0.08 (0.22) 0.10 (0.1)
S. America % 0.05 (0.16) 0.06 (0.17) 0.04 (0.10) 0.02 (0.0)
Australia % 0.12 (0.29) 0.13 (0.30) 0.08 (0.24) 0.05 (0.1)
N 128 106 22 128
Notes: Table A2 presents an alternative calculation to the geographic exposure shown in Table 1.1 Panel B.
Each variable is the bank’s percentage exposure to a geographic exposure, calculated as the credit extended
to each geography over the bank’s total lending. "Not Failed" and "Failed" refers to whether a bank
suspended or closed during the crisis. Means are reported first, and standard deviations are given in
parentheses. "Diff" refers to the difference in means between groups. Standard errors are reported in
parentheses for the "Diff" column. Significance is marked by *p < 0.1, **p < 0.05, ***p < 0.01.
143
Table A3: Robustness to removing cotton exporting countries: immediate effect of exposure to bank
failures on port-level shipping
All excl USA excl Brazil excl Egypt excl all cotton
(1) (2) (3) (4) (5)
Failpo ⇥ post -0.531*** -0.524*** -0.533*** -0.488*** -0.485***
[0.171] [0.175] [0.181] [0.167] [0.184]
Capital city ⇥ post Y Y Y Y Y
Age of banks ⇥ post Y Y Y Y Y
# non-Brit banks ⇥ post Y Y Y Y Y
Fraction to UK ⇥ post Y Y Y Y Y
Countryo ⇥ post FE Y Y Y Y Y
Portp FE Y Y Y Y Y
N 578 560 556 564 524
Ports 289 280 278 282 262
Clusters 54 53 53 53 51
Notes: Table A3 reports estimates from the difference-in-difference regressions from the two-period panel of
port-level shipping activity in the year before and after the crisis. The dependent variable is the ln of the
number of ships departing from each port. Fail po is the share of the port’s British banks that failed during
the crisis. post is a dummy for the post-crisis year that takes the value of 1 after May 1866 and 0 otherwise.
The time-invariant control variables are measured in 1865 and interacted with the post dummy. In columns
2–4, ports from the United States, Brazil, and Egypt are excluded respectively. In column 5, ports from all
three cotton exporting countries are excluded. Standard errors in brackets are clustered by country of
origin. *p < 0.1, **p < 0.05, ***p < 0.01
144
Table A4: Robustness to controls: immediate effect of exposure to bank failures on country-level shipping
Panel A: Industry composition of exports
(1) (2) (3) (4) (5) (6) (7)

Failo ⇥ post -0.498** -0.584** -0.523** -0.505** -0.519** -0.555** -0.582**
[0.224] [0.240] [0.220] [0.234] [0.224] [0.221] [0.242]
non-Brit banks ⇥ post Y
ln(sugar) ⇥ post Y
ln(cotton raw) ⇥ post Y
ln(cotton manu) ⇥ post Y
ln(grains) ⇥ post Y
ln(tobacco) ⇥ post Y
ln(coffee) ⇥ post Y
Country FE Y Y Y Y Y Y Y
N 108 108 108 108 108 108 108
Clusters 54 54 54 54 54 54 54
Panel B: Monetary standard and conflict
(1) (2) (3) (4) (5) (6) (7)

Failo ⇥ post -0.507** -0.501** -0.581** -0.595** -0.512** -0.501** -0.464**
[0.227] [0.231] [0.248] [0.249] [0.208] [0.238] [0.228]
Size ⇥ post Y
Gold ⇥ post Y
Silver ⇥ post Y
Bimetallic ⇥ post Y
Conflict, any ⇥ post Y
Conflict, interstate ⇥ post Y
Conflict, other ⇥ post Y
Country FE Y Y Y Y Y Y Y
N 108 106 106 106 108 108 108
Clusters 54 53 53 53 54 54 54
Notes: Table A4 reports estimates from the difference-in-difference regressions from the two-period panel of
country-level shipping activity in the year before and after the crisis. The dependent variable is the ln of
the number of ships departing from each country. Failo is the share of the country’s banks that failed
during the crisis. The mean of Failo is 0.11, and the standard deviation is 0.17. post is a dummy for the
post-crisis year that takes the value of 1 after May 1866 and 0 otherwise. The time-invariant control
variables are measured in 1865 and interacted with the post dummy. In Panel A, they include the ln values
of sugar, raw cotton, cotton manufactured goods, grains, tobacco, and coffee exports. The ln values of
industry exports are replaced with 0 if the country does not export those products. In Panel B, they include
the size of the country proxied by the total value of exports, the monetary standard of the country, and
engagement in conflict. Controls are added sequentially and the coefficients are stable. Standard errors in
brackets are clustered by country of origin. *p < 0.1, **p < 0.05, ***p < 0.01
145
Table A5: Robustness to different clustering: immediate effect of exposure to bank failures on
country-level exports

(1) (2) (3) (4) (5)
b1865 -0.240 -0.240 -0.240 -0.240 -0.240
[0.214] [0.151] [0.192] [0.176] [0.318]
b1867 -0.921 -0.921 -0.921 -0.921 -0.921***
[0.603] [0.643] [0.575] [0.661] [0.0689]
b1868 -1.611*** -1.611*** -1.611*** -1.611** -1.611***
[0.551] [0.585] [0.506] [0.616] [0.393]
b1869 -1.872*** -1.872*** -1.872*** -1.872*** -1.872***
[0.410] [0.584] [0.493] [0.519] [0.263]
b1870 -1.633*** -1.633** -1.633*** -1.633*** -1.633***
[0.434] [0.621] [0.536] [0.540] [0.293]
Controls Y Y Y Y Y
Countryo FE Y Y Y Y Y
I(Brit bankot ) Y Y Y Y Y
Countrydt Y Y Y Y Y
N 2952 2952 2952 2952 2952
Clustering Orig country Dest country Orig-Dest pair Multi: Orig, Dest Multi: Orig, Dest, year
Adj. R2 0.551 0.550 0.551 0.550 0.550
Notes: Table A5 reports estimates from the annual dynamic difference-in-difference regressions from the
panel of country-level values of trade. The dependent variable is the ln value of exports from origin
country o to destination country d. There are 83 exporting countries from 1865-1870. Failo is the share of
the country’s banks that failed. post is a dummy for the post-crisis years 1867-1870. Baseline controls are
the log distance between country o and country d. Standard errors in brackets are clustered according to
the row labeled “Clustering.” *p < 0.1, **p < 0.05, ***p < 0.01
146
Table A6: Elasticity of trade to physical distance
ln(EXodt ) = qln(distance)od + got + gdt + G0 Xodt + # odt

(1) (2) (3) (4) (5)
log distance od -1.116*** -1.021*** -0.982*** -1.194*** -1.037***
[0.0851] [0.0910] [0.101] [0.0856] [0.101]
Countryot FE Y Y Y Y Y
Countrydt FE Y Y Y Y Y
Common language ⇥ t Y Y
Common border ⇥ t Y Y
Common empire ⇥ t Y Y
N 67378 67378 67378 67378 67378
Clusters 119 119 119 119 119
Adj. R2 0.530 0.548 0.534 0.559 0.564
Notes: Table A6 reports estimates for q, the elasticity of trade to physical distance, from the above
estimation. All specifications are estimated using the full panel of bilateral trade data from 1850–1914. The
baseline specification is given in Column 1. Columns 2–5 control for standard gravity measurements of
bilateral resistance. The dependent variable is the ln value of exports from origin country o to destination
country d. The origin country-year fixed effects effectively drop the countries that only appear in the trade
data for one year. There are 10 such countries and therefore only 119 clusters. Standard errors in brackets
are clustered by origin country. *p < 0.1, **p < 0.05, ***p < 0.01.
147
Table A7: Long-term effects of financing shock on country-level exports
ln(Exodt ) = b t Fo + G0 Xot + go + gdt + qt ln(dist)od + # odt

(1) (2) (3) (4) (5) (6) (7) (8)
b1850-1855 0.0309 -0.248 -0.217 -0.171 -0.180 0.311 -0.284 -0.254
[0.217] [0.330] [0.413] [0.343] [0.344] [0.486] [0.306] [0.296]
b1856-1860 -0.0624 -0.152 -0.324 0.0465 -0.0696 -0.155 -0.0873 -0.0469
[0.145] [0.224] [0.316] [0.192] [0.229] [0.350] [0.160] [0.141]
b1866-1870 -1.528*** -1.476*** -1.513*** -1.543*** -1.433*** -2.160*** -1.569*** -1.424***
[0.322] [0.372] [0.407] [0.415] [0.410] [0.579] [0.577] [0.524]
b1871-1875 -1.772*** -1.743*** -1.841*** -1.618*** -1.651*** -2.206** -1.587** -1.713***
[0.462] [0.522] [0.575] [0.556] [0.526] [0.851] [0.598] [0.524]
b1876-1880 -1.902*** -1.745*** -1.963*** -1.538*** -1.623*** -2.320** -1.445** -1.626***
[0.521] [0.568] [0.651] [0.564] [0.557] [0.891] [0.568] [0.515]
b1881-1885 -1.483*** -1.347** -1.475** -1.221** -1.221** -2.160** -1.200** -1.320***
[0.449] [0.547] [0.653] [0.553] [0.542] [0.906] [0.520] [0.473]
b1886-1890 -1.394*** -1.199** -1.437** -1.095** -1.117** -1.895** -1.249*** -1.381***
[0.390] [0.506] [0.621] [0.523] [0.526] [0.858] [0.453] [0.429]
b1891-1895 -1.319*** -1.148** -1.457** -0.979* -1.008* -1.887** -1.346** -1.267**
[0.383] [0.511] [0.645] [0.527] [0.505] [0.736] [0.519] [0.508]
b1896-1900 -1.391*** -1.210*** -1.489** -1.041** -1.188** -1.956*** -1.468*** -1.481***
[0.325] [0.451] [0.611] [0.481] [0.459] [0.648] [0.415] [0.420]
b1901-1905 -1.046** -0.848 -1.256* -0.530 -0.993* -1.523** -1.090** -1.055**
[0.403] [0.535] [0.723] [0.514] [0.523] [0.699] [0.489] [0.510]
b1906-1910 -0.877** -0.705 -1.104 -0.412 -0.846 -1.249 -0.891** -1.117***
[0.424] [0.523] [0.687] [0.504] [0.558] [0.810] [0.377] [0.391]
b1911-1914 -1.009* -0.868 -1.234* -0.583 -0.972 -1.093 -0.815* -1.417***
[0.521] [0.573] [0.731] [0.551] [0.624] [0.846] [0.452] [0.429]
Controls Y Y Y Y Y Y Y Y
Countryo FE Y Y Y Y Y Y Y Y
ln(cottono ) ⇥ t Y
ln(cotton manuo ) ⇥ t Y
ln(populationo ) ⇥ Y
SITC industryo ⇥ t Y
Regiono ⇥ t Y Y
I(Brit banko ) ⇥ t Y Y Y Y Y Y Y Y
Countryd Y
Countrydt Y Y Y Y Y Y Y
N 67378 67378 67378 67378 55391 49006 49006 67378
Clusters 129 129 129 129 54 48 48 129
Adj. R2 0.530 0.530 0.531 0.531 0.545 0.559 0.558 0.532
Notes: Table A7 reports the point estimates for the long-term effects of the credit shock on the value of
country-level exports. The dependent variable is the log value of exports from origin country o to
destination country d. Baseline controls are the log distance between country o and country d. Cotton,
cotton manufactured goods, and population are calculated in 1865 and interacted with the 5-year dummies.
Countries that did not export cotton are given ln values of zero. Controlling for pre-crisis population and
the SITC industry of exports reduces the sample size to countries that were exporting pre-crisis. Column 7
artificially restricts the sample to countries with SITC codes available. Standard errors in brackets are
clustered by the origin country. *p < 0.1, **p < 0.05, ***p < 0.01
148
Table A8: Effect of bank failures from 1850–2014

(1) (2) (3)
b1850-1855 0.202 [0.306] -0.256 [0.356] -0.117 [0.357]
b1856-1860 -0.184 [0.192] -0.317 [0.302] -0.199 [0.299]
b1866-1870 -2.430*** [0.837] -1.985*** [0.595] -1.752*** [0.598]
b1871-1875 -2.793*** [0.984] -2.386*** [0.847] -2.090** [0.838]
b1876-1880 -2.779*** [0.993] -2.248** [0.869] -1.928** [0.853]
b1881-1885 -2.429*** [0.828] -1.789** [0.771] -1.480* [0.756]
b1886-1890 -2.367*** [0.662] -1.689** [0.678] -1.428** [0.678]
b1891-1895 -2.369*** [0.688] -1.694** [0.700] -1.376** [0.671]
b1896-1900 -2.432*** [0.558] -1.800*** [0.607] -1.579*** [0.585]
b1901-1905 -2.163*** [0.600] -1.451** [0.700] -1.367* [0.690]
b1906-1910 -1.982*** [0.533] -1.270* [0.645] -1.196* [0.665]
b1911-1915 -2.276*** [0.531] -1.519** [0.634] -1.447** [0.669]
b1916-1920 -3.666*** [1.009] -2.859*** [1.088] -2.720** [1.121]
b1921-1925 -2.487*** [0.637] -1.755** [0.739] -1.833** [0.793]
b1926-1930 -2.010*** [0.601] -1.433** [0.700] -1.530** [0.741]
b1931-1935 -1.598** [0.628] -1.031 [0.747] -1.281 [0.786]
b1936-1940 -1.725** [0.688] -1.061 [0.784] -1.323 [0.829]
b1941-1945 -2.925* [1.588] -2.291 [1.386] -2.798** [1.145]
b1946-1950 -1.752** [0.752] -1.288 [0.803] -1.625** [0.776]
b1951-1955 -1.934*** [0.669] -1.404* [0.739] -1.643** [0.764]
b1956-1960 -2.010*** [0.654] -1.488** [0.719] -1.727** [0.748]
b1961-1965 -2.102*** [0.624] -1.558** [0.705] -1.774** [0.749]
b1966-1970 -1.799*** [0.632] -1.240* [0.735] -1.568* [0.782]
b1971-1975 -1.461** [0.680] -0.848 [0.806] -1.246 [0.858]
b1976-1980 -1.402** [0.669] -0.762 [0.803] -1.167 [0.843]
b1981-1985 -1.512** [0.688] -0.891 [0.818] -1.344 [0.865]
b1986-1990 -1.353* [0.694] -0.735 [0.832] -1.306 [0.875]
b1991-1995 -1.756** [0.691] -1.145 [0.832] -1.724* [0.881]
b1996-2000 -1.755** [0.686] -1.134 [0.817] -1.882** [0.835]
b2001-2005 -1.630** [0.720] -1.020 [0.841] -1.866** [0.835]
b2006-2010 -1.537** [0.750] -0.933 [0.858] -1.878** [0.809]
b2011-2014 -1.413* [0.773] -0.806 [0.869] -1.700** [0.816]
Controls Y Y Y
Countryo FE Y Y Y
I(Brit banko ) ⇥ t Y Y Y
ln(populationo ) ⇥ t Y
Countryd Y
Countrydt Y Y
N 665866 665866 414777
Clusters 137 137 54
Adj. R2 0.654 0.680 0.748
Notes: Table A8 reports the coefficients every five years. The control variables are the same as defined in
Table A7. Standard errors in brackets are clustered by the origin country. *p < 0.1, **p < 0.05, ***p < 0.01
149
Table A9: Long-term effects: robustness to gravity measures of commonality
ln(Exodt ) = b t Fo + G0 Xodt + go + gdt + qt ln(dist)od + # odt

(1) (2) (3) (4) (5) (6)
b1850-1855 -0.340 -0.259 -0.227 -0.324 -0.264 -0.273
[0.322] [0.306] [0.315] [0.334] [0.324] [0.316]
b1856-1860 -0.203 -0.139 -0.186 -0.157 -0.139 -0.235
[0.234] [0.221] [0.239] [0.228] [0.226] [0.239]
b1866-1870 -1.452*** -1.537*** -1.163*** -1.491*** -1.535*** -1.221***
[0.329] [0.342] [0.391] [0.306] [0.347] [0.379]
b1871-1875 -1.732*** -1.797*** -1.383** -1.754*** -1.797*** -1.446**
[0.502] [0.522] [0.604] [0.460] [0.522] [0.595]
b1876-1880 -1.709*** -1.803*** -1.407** -1.740*** -1.802*** -1.479**
[0.558] [0.571] [0.656] [0.536] [0.570] [0.653]
b1881-1885 -1.290** -1.393** -1.033* -1.317** -1.391** -1.102*
[0.547] [0.556] [0.602] [0.544] [0.556] [0.595]
b1886-1890 -1.113** -1.236** -0.869 -1.172** -1.236** -0.933*
[0.496] [0.499] [0.526] [0.500] [0.499] [0.527]
b1891-1895 -1.079** -1.203** -0.847 -1.107** -1.203** -0.906*
[0.490] [0.503] [0.538] [0.491] [0.504] [0.535]
b1896-1900 -1.154*** -1.269*** -0.905* -1.223*** -1.269*** -0.961**
[0.423] [0.429] [0.468] [0.425] [0.432] [0.466]
b1901-1905 -0.789 -0.888* -0.587 -0.859* -0.885* -0.645
[0.505] [0.509] [0.569] [0.498] [0.512] [0.566]
b1906-1910 -0.640 -0.735 -0.437 -0.724 -0.738 -0.493
[0.505] [0.503] [0.549] [0.503] [0.504] [0.551]
b1911-1914 -0.842 -0.893 -0.610 -0.939 -0.900 -0.669
[0.568] [0.554] [0.597] [0.572] [0.553] [0.601]
Common language 1.102***
[0.157]
Common border 0.854***
[0.212]
Common empire 1.741***
[0.162]
Common language ⇥ t Y
Common border ⇥ t Y
Common empire ⇥ t Y
I(Brit banko ) ⇥ t Y Y Y Y Y Y
Countrydt Y Y Y Y Y Y
N 67378 67378 67378 67378 67378 67378
Clusters 129 129 129 129 129 129
Adj. R2 0.547 0.534 0.557 0.547 0.533 0.557
Notes: Table A9 reports the coefficients every five years. The control variables are time-invariant and
time-varying measures of distance standard to gravity estimations, such as common language. Standard
errors in brackets are clustered by the origin country. *p < 0.1, **p < 0.05, ***p < 0.01
150
Table A10: Long-term effects: robustness to monetary standard and conflict

(1) (2) (3) (4) (5) (6) (7)
b1850-1855 -0.114 -0.280 -0.249 -0.260 -0.261 -0.243 -0.245
[0.430] [0.321] [0.377] [0.432] [0.336] [0.331] [0.342]
b1856-1860 -0.308 -0.177 -0.110 0.0321 -0.0357 -0.141 -0.139
[0.280] [0.228] [0.267] [0.283] [0.197] [0.232] [0.219]
b1866-1870 -1.527*** -1.581*** -1.693*** -1.488*** -1.368*** -1.472*** -1.460***
[0.426] [0.382] [0.384] [0.410] [0.427] [0.369] [0.373]
b1871-1875 -1.865*** -1.887*** -1.992*** -1.745*** -1.553*** -1.735*** -1.726***
[0.626] [0.510] [0.486] [0.535] [0.579] [0.523] [0.524]
b1876-1880 -1.966*** -1.875*** -1.992*** -1.739*** -1.550** -1.738*** -1.727***
[0.690] [0.562] [0.555] [0.615] [0.594] [0.572] [0.569]
b1881-1885 -1.520** -1.472*** -1.595*** -1.341** -1.152** -1.344** -1.330**
[0.680] [0.548] [0.608] [0.651] [0.556] [0.551] [0.547]
b1886-1890 -1.337** -1.226** -1.444** -1.190* -1.005* -1.193** -1.182**
[0.663] [0.538] [0.592] [0.614] [0.514] [0.511] [0.503]
b1891-1895 -1.342* -1.142** -1.407** -1.164* -0.956* -1.141** -1.130**
[0.684] [0.545] [0.582] [0.625] [0.534] [0.521] [0.512]
b1896-1900 -1.386** -1.196** -1.448*** -1.200** -1.019** -1.202** -1.193***
[0.629] [0.466] [0.530] [0.558] [0.483] [0.461] [0.453]
b1901-1905 -0.998 -0.789 -1.048* -0.839 -0.683 -0.842 -0.831
[0.736] [0.549] [0.590] [0.637] [0.531] [0.545] [0.537]
b1906-1910 -0.840 -0.651 -0.924 -0.671 -0.521 -0.701 -0.686
[0.717] [0.541] [0.615] [0.622] [0.525] [0.528] [0.523]
b1911-1914 -0.958 -0.795 -1.126* -0.875 -0.668 -0.863 -0.853
[0.741] [0.596] [0.658] [0.665] [0.564] [0.577] [0.573]
Controls Y Y Y Y Y Y Y
Countryo FE Y Y Y Y Y Y Y
Gold standardo ⇥ t Y
Silver standardo ⇥ t Y
Conflict (any)o ⇥ t Y
Conflict (interstate)o ⇥ t Y
Conflict (other)o ⇥ t Y
Countryot war Y
Country-pairodt war Y
I(Brit banko ) ⇥ t Y Y Y Y Y Y Y
N 56937 56937 67378 67378 67378 67378 67378
Clusters 55 55 129 129 129 129 129
Adj. R2 0.543 0.543 0.530 0.530 0.530 0.530 0.530
Notes: Table A10 reports the coefficients every five years. The monetary and conflict variables are binary
variables taking a value of 1 if the exporting country had that characteristic in 1865 or 1866 and are
interacted with year dummies. Column 6 controls for war in the origin country (including civil war) in any
year, and Column 7 controls for war between dyadic pairs of countries in any year. Standard errors in
brackets are clustered by the origin country. *p < 0.1, **p < 0.05, ***p < 0.01
151
Table A11: Long-term effects: robustness to industry composition of exports

(1) (2) (3) (4) (5) (6)
b1850-1855 -0.292 -0.341 -0.231 -0.327 -0.379 -0.349
[0.314] [0.402] [0.344] [0.398] [0.386] [0.319]
b1856-1860 -0.141 -0.362 -0.103 -0.335 -0.207 -0.256
[0.226] [0.249] [0.233] [0.250] [0.233] [0.229]
b1866-1870 -1.381*** -1.632*** -1.620*** -1.501*** -1.736*** -1.469***
[0.360] [0.441] [0.388] [0.478] [0.426] [0.445]
b1871-1875 -1.518*** -2.030*** -1.789*** -1.752*** -2.024*** -1.814***
[0.476] [0.604] [0.515] [0.642] [0.557] [0.511]
b1876-1880 -1.551*** -2.155*** -1.760*** -1.751** -2.038*** -1.800***
[0.531] [0.667] [0.536] [0.701] [0.589] [0.547]
b1881-1885 -1.185** -1.798*** -1.393** -1.390** -1.676*** -1.421***
[0.529] [0.684] [0.557] [0.678] [0.602] [0.525]
b1886-1890 -1.097** -1.688** -1.306*** -1.314** -1.561*** -1.212**
[0.518] [0.673] [0.499] [0.620] [0.573] [0.479]
b1891-1895 -1.066** -1.639** -1.210** -1.239* -1.444** -1.184**
[0.519] [0.683] [0.479] [0.630] [0.578] [0.495]
b1896-1900 -1.168** -1.696*** -1.146*** -1.207** -1.480*** -1.294***
[0.453] [0.620] [0.438] [0.562] [0.526] [0.420]
b1901-1905 -0.841 -1.422** -0.702 -0.706 -1.011* -0.880
[0.524] [0.695] [0.486] [0.615] [0.546] [0.544]
b1906-1910 -0.820 -1.292* -0.537 -0.545 -0.871 -0.835
[0.508] [0.673] [0.499] [0.633] [0.600] [0.515]
b1911-1914 -0.964* -1.442** -0.670 -0.665 -1.051 -0.937
[0.532] [0.698] [0.558] [0.700] [0.679] [0.587]
ln(coffeeo ) ⇥ t Y
ln(grainso ) ⇥ t Y
ln(bulliono ) ⇥ t Y
ln(alcoholo ) ⇥ t Y
ln(tobaccoo ) ⇥ t Y
Commodities shareo ⇥ t Y
N 67378 67378 67378 67378 67378 62109
Clusters 129 129 129 129 129 81
Adj. R2 0.531 0.531 0.531 0.531 0.532 0.538
Notes: Table A11 reports the coefficients every five years. The industry-level exports are calculated in 1865
and interacted with the 5-year dummies. Countries that did not export a commodity are given ln values of
zero. The Commodities share of exports is the fraction of goods exported in 1865 that are categorized as
raw or primary products. Standard errors in brackets are clustered by the origin country. *p < 0.1, **p <
0.05, ***p < 0.01
152
Table A12: Long-term effects: robustness to excluding cotton exporting countries

(1) (2) (3) (4) (5) (6)
b1850-1855 0.0141 -0.281 0.0470 -0.231 0.0783 -0.167
[0.234] [0.358] [0.221] [0.328] [0.201] [0.311]
b1856-1860 -0.0175 0.0118 -0.0715 -0.162 -0.0354 -0.104
[0.153] [0.214] [0.148] [0.227] [0.143] [0.216]
b1866-1870 -1.590*** -1.498*** -1.501*** -1.469*** -1.453*** -1.373***
[0.347] [0.439] [0.326] [0.377] [0.294] [0.371]
b1871-1875 -1.715*** -1.550** -1.744*** -1.737*** -1.710*** -1.656***
[0.498] [0.606] [0.468] [0.524] [0.426] [0.505]
b1876-1880 -1.789*** -1.439** -1.889*** -1.758*** -1.843*** -1.654***
[0.556] [0.609] [0.529] [0.571] [0.488] [0.550]
b1881-1885 -1.374*** -1.068* -1.498*** -1.374** -1.434*** -1.260**
[0.483] [0.588] [0.456] [0.551] [0.435] [0.534]
b1886-1890 -1.330*** -0.940* -1.398*** -1.218** -1.342*** -1.117**
[0.427] [0.561] [0.395] [0.510] [0.387] [0.497]
b1891-1895 -1.233*** -0.881 -1.325*** -1.174** -1.270*** -1.067**
[0.429] [0.573] [0.390] [0.515] [0.373] [0.502]
b1896-1900 -1.319*** -0.966* -1.393*** -1.227*** -1.334*** -1.121**
[0.363] [0.530] [0.332] [0.456] [0.325] [0.444]
b1901-1905 -0.903** -0.481 -1.038** -0.860 -0.971** -0.745
[0.420] [0.559] [0.412] [0.541] [0.400] [0.527]
b1906-1910 -0.750* -0.366 -0.863** -0.711 -0.796* -0.592
[0.437] [0.552] [0.429] [0.528] [0.425] [0.508]
b1911-1914 -0.883* -0.535 -1.041** -0.928 -0.917* -0.745
[0.528] [0.592] [0.526] [0.574] [0.523] [0.555]
excluding USA Y Y
excluding Brazil Y Y
excluding Egypt Y Y
Countryd Y Y Y
Countrydt Y Y Y
N 63851 63851 66381 66381 66570 66570
Clusters 128 128 128 128 128 128
Adj. R2 0.524 0.524 0.531 0.531 0.530 0.530
Notes: Table A12 reports the coefficients every five years. Exports from the USA, Brazil, and Egypt are
excluded in columns 1–2, 3–4, and 5–6, respectively. Standard errors in brackets are clustered by the origin
country. *p < 0.1, **p < 0.05, ***p < 0.01
153
Table A13: Long-term effects: robustness to contemporaneous financial crises

(1) (2) (3) (4) (5) (6) (7) (8)
b1850-1855 -0.298 -0.288 -0.247 -0.297 -0.288 -0.291 -0.288 -0.270
[0.333] [0.342] [0.324] [0.336] [0.335] [0.336] [0.332] [0.327]
b1856-1860 -0.152 -0.134 -0.118 -0.153 -0.140 -0.140 -0.115 -0.103
[0.228] [0.219] [0.212] [0.228] [0.227] [0.227] [0.227] [0.226]
b1866-1870 -1.410*** -1.397*** -1.437*** -1.411*** -1.428*** -1.422*** -1.419*** -1.443***
[0.401] [0.405] [0.397] [0.401] [0.404] [0.404] [0.407] [0.397]
b1871-1875 -1.724*** -1.701*** -1.757*** -1.741*** -1.761*** -1.725*** -1.724*** -1.707***
[0.538] [0.537] [0.545] [0.540] [0.546] [0.541] [0.542] [0.549]
b1876-1880 -1.713*** -1.697*** -1.717*** -1.739*** -1.732*** -1.685*** -1.712*** -1.683***
[0.581] [0.574] [0.574] [0.587] [0.595] [0.580] [0.583] [0.589]
b1881-1885 -1.242** -1.217** -1.278** -1.256** -1.260** -1.231** -1.260** -1.262**
[0.553] [0.548] [0.542] [0.553] [0.554] [0.549] [0.553] [0.551]
b1886-1890 -1.126** -1.102** -1.188** -1.129** -1.161** -1.147** -1.127** -1.127**
[0.521] [0.515] [0.507] [0.520] [0.515] [0.516] [0.521] [0.518]
b1891-1895 -1.084** -1.058** -1.140** -1.086** -1.109** -1.097** -1.060** -1.044*
[0.534] [0.525] [0.530] [0.531] [0.524] [0.526] [0.529] [0.533]
b1896-1900 -1.229** -1.210** -1.248*** -1.232** -1.243*** -1.234** -1.234** -1.229**
[0.469] [0.464] [0.462] [0.468] [0.462] [0.464] [0.471] [0.474]
b1901-1905 -1.038* -1.014* -0.971* -1.042* -1.057* -1.046* -1.033* -1.031*
[0.549] [0.540] [0.549] [0.548] [0.538] [0.540] [0.550] [0.552]
b1906-1910 -0.832 -0.806 -0.748 -0.836 -0.868 -0.853 -0.836 -0.825
[0.552] [0.542] [0.527] [0.551] [0.538] [0.539] [0.553] [0.551]
b1911-1914 -0.920 -0.898 -0.839 -0.923 -0.956 -0.941 -0.926 -0.915
[0.594] [0.585] [0.560] [0.593] [0.578] [0.580] [0.591] [0.589]
Controls Y Y Y Y Y Y Y Y
Countryo FE Y Y Y Y Y Y Y Y
Currency crisisot Y
Inflation crisisot Y
Stock mkt crisisot Y
Sovereign debt (domestic)ot Y
Sovereign debt (external)ot Y
Sovereign debt (any)ot Y
Banking crisisot Y
Any crisisot Y
I(Brit banko ) ⇥ t Y Y Y Y Y Y Y Y
Countrydt Y Y Y Y Y Y Y Y
N 57305 57305 52480 57305 57305 57305 57305 57305
Clusters 62 62 44 62 62 62 62 62
Adj. R2 0.543 0.544 0.545 0.543 0.544 0.544 0.544 0.544
Notes: Table A13 reports the coefficients every five years. Different types of financial crises are binary
variables, which take the value of 1 if the exporting country is experiencing it in any given year. These are
contemporaneous measures taken from Reinhart and Rogoff (2009b). Data limitations reduce the number
of observations. Standard errors in brackets are clustered by the origin country. *p < 0.1, **p < 0.05, ***p <
0.01
154
Table A14: Long-term effects: robustness to financial crises in 1865

(1) (2) (3) (4) (5)
b1850-1855 -0.202 -0.314 -0.264 -0.306 -0.187
[0.325] [0.359] [0.330] [0.342] [0.356]
b1856-1860 -0.0976 -0.203 -0.226 -0.234 -0.257
[0.214] [0.264] [0.240] [0.229] [0.245]
b1866-1870 -1.352*** -1.541*** -1.248*** -1.419*** -1.340***
[0.431] [0.399] [0.417] [0.406] [0.399]
b1871-1875 -1.621*** -1.941*** -1.634*** -1.769*** -1.657***
[0.538] [0.432] [0.548] [0.524] [0.504]
b1876-1880 -1.608*** -1.958*** -1.701*** -1.754*** -1.642***
[0.566] [0.446] [0.588] [0.572] [0.535]
b1881-1885 -1.179** -1.465*** -1.260** -1.253** -1.178**
[0.546] [0.522] [0.551] [0.553] [0.547]
b1886-1890 -1.063** -1.320** -1.111** -1.120** -1.074*
[0.526] [0.505] [0.531] [0.523] [0.538]
b1891-1895 -0.958* -1.281** -1.125** -1.069* -1.024*
[0.527] [0.502] [0.524] [0.535] [0.535]
b1896-1900 -1.110** -1.386*** -1.247** -1.229** -1.178**
[0.473] [0.436] [0.474] [0.471] [0.482]
b1901-1905 -0.883 -1.129** -1.140** -1.030* -0.969*
[0.542] [0.522] [0.544] [0.549] [0.537]
b1906-1910 -0.640 -0.909* -0.933 -0.827 -0.756
[0.540] [0.526] [0.565] [0.553] [0.557]
b1911-1914 -0.726 -0.987* -0.990 -0.922 -0.852
[0.583] [0.582] [0.607] [0.592] [0.604]
Controls Y Y Y Y Y
Countryo FE Y Y Y Y Y
Inflation crisiso ⇥ t Y
Stock mkt crisiso ⇥ t Y
Sovereign debt crisiso ⇥ t Y
Banking crisiso ⇥ t Y
Any crisiso ⇥ t Y
I(Brit banko ) ⇥ t Y Y Y Y Y
Countrydt Y Y Y Y Y
N 57305 52483 57305 57305 57305
Clusters 62 44 62 62 62
Adj. R2 0.544 0.545 0.544 0.543 0.543
Notes: Table A14 reports the coefficients every five years. Different types of financial crises are binary
variables, which take the value of 1 if the exporting country is experiencing it in 1865, taken from Reinhart
and Rogoff (2009b), and interacted with year dummies. No country experienced a currency crisis or
domestic sovereign debt crisis in 1865 so these are not reported. Standard errors in brackets are clustered
by the origin country. *p < 0.1, **p < 0.05, ***p < 0.01
155
Table A15: Long-term effects: robustness to borrowing from London Stock Exchange

(1) (2) (3) (4) (5) (6)
b1850-1855 -0.249 -0.251 -0.251 -0.247 -0.248 -0.255
[0.329] [0.330] [0.331] [0.330] [0.331] [0.331]
b1856-1860 -0.152 -0.150 -0.155 -0.152 -0.150 -0.151
[0.225] [0.224] [0.225] [0.224] [0.220] [0.224]
b1866-1870 -1.495*** -1.470*** -1.415*** -1.385*** -1.433*** -1.457***
[0.391] [0.384] [0.382] [0.395] [0.374] [0.372]
b1871-1875 -1.775*** -1.785*** -1.712*** -1.671*** -1.751*** -1.723***
[0.548] [0.546] [0.538] [0.541] [0.532] [0.526]
b1876-1880 -1.808*** -1.817*** -1.752*** -1.715*** -1.744*** -1.792***
[0.586] [0.583] [0.567] [0.577] [0.573] [0.571]
b1881-1885 -1.376** -1.469*** -1.334** -1.373** -1.321** -1.375**
[0.547] [0.549] [0.532] [0.538] [0.564] [0.545]
b1886-1890 -1.231** -1.229** -1.168** -1.193** -1.192** -1.220**
[0.503] [0.508] [0.500] [0.506] [0.512] [0.503]
b1891-1895 -1.180** -1.117** -1.133** -1.138** -1.135** -1.161**
[0.510] [0.509] [0.510] [0.513] [0.511] [0.508]
b1896-1900 -1.119** -1.127** -1.172*** -1.219*** -1.206*** -1.257***
[0.465] [0.445] [0.439] [0.439] [0.451] [0.447]
b1901-1905 -0.596 -0.714 -0.839 -0.934* -0.846 -0.914*
[0.536] [0.513] [0.537] [0.552] [0.558] [0.532]
b1906-1910 -0.564 -0.789 -0.718 -0.727 -0.722 -0.748
[0.569] [0.524] [0.516] [0.532] [0.549] [0.520]
b1911-1914 -0.811 -1.000* -0.921* -0.904 -0.915 -0.916
[0.616] [0.573] [0.547] [0.569] [0.578] [0.569]
Sovereign debtot Y
Any equityot Y
Corporate debtot Y
Railway issuanceot Y
Bank issuanceot Y
Any industry issuanceot Y
N 67378 67378 67378 67378 67378 67378
Clusters 129 129 129 129 129 129
Adj. R2 0.531 0.531 0.530 0.530 0.530 0.530
Notes: Table A15 reports the coefficients every five years. Variables denoting borrowing on the London
Stock Exchange are binary variables which take the value of 1 if the exporting country issued a given type
of debt or equity each year. These data are taken from the Investor’s Manual Monthly, discussed in
Appendix 3.3. Standard errors in brackets are clustered by the origin country. *p < 0.1, **p < 0.05, ***p <
0.01
156
Table A16: Long-term effects: robustness to composition of banks

(1) (2) (3) (4) (5) (6) (7)
b1850-1855 -0.141 -0.271 -0.173 -0.258 -0.159 -0.129 -0.215
[0.350] [0.335] [0.341] [0.327] [0.349] [0.321] [0.322]
b1856-1860 -0.0169 -0.0179 -0.0488 -0.136 -0.0133 0.0231 0.121
[0.250] [0.259] [0.221] [0.220] [0.261] [0.209] [0.234]
b1866-1870 -1.361*** -1.739*** -1.575*** -1.604*** -1.448*** -1.669*** -1.762***
[0.421] [0.435] [0.403] [0.414] [0.398] [0.421] [0.437]
b1871-1875 -1.554** -2.026*** -1.851*** -1.900*** -1.544*** -1.962*** -2.005***
[0.636] [0.541] [0.550] [0.570] [0.573] [0.567] [0.549]
b1876-1880 -1.641*** -2.159*** -1.963*** -2.018*** -1.530*** -2.105*** -2.169***
[0.582] [0.602] [0.631] [0.594] [0.555] [0.600] [0.593]
b1881-1885 -1.253** -1.722*** -1.616** -1.640*** -1.191** -1.763*** -1.790***
[0.558] [0.627] [0.632] [0.587] [0.534] [0.614] [0.629]
b1886-1890 -1.154** -1.589*** -1.491** -1.482** -0.995** -1.632*** -1.686***
[0.497] [0.602] [0.618] [0.574] [0.493] [0.609] [0.606]
b1891-1895 -1.091** -1.501** -1.450** -1.370** -0.934* -1.561** -1.565***
[0.506] [0.588] [0.641] [0.553] [0.504] [0.603] [0.596]
b1896-1900 -1.155*** -1.540*** -1.510** -1.392*** -0.942** -1.568*** -1.635***
[0.440] [0.527] [0.583] [0.483] [0.460] [0.536] [0.537]
b1901-1905 -0.783 -1.308** -1.166* -1.055* -0.644 -1.252** -1.406**
[0.514] [0.548] [0.675] [0.537] [0.494] [0.600] [0.555]
b1906-1910 -0.634 -1.190** -1.028 -0.904 -0.519 -1.072* -1.278**
[0.503] [0.563] [0.661] [0.550] [0.512] [0.599] [0.572]
b1911-1914 -0.788 -1.348** -1.207* -1.070* -0.690 -1.261* -1.448**
[0.554] [0.620] [0.716] [0.605] [0.543] [0.657] [0.637]
Controls Y Y Y Y Y Y Y
Countryo FE Y Y Y Y Y Y Y
British banksot Y
Local banksot Y
French banksot Y
German banksot Y
US banksot Y
European (non-Brit) banksot Y
Total banksot Y
I(Brit banko ) ⇥ t Y Y Y Y Y Y Y
N 67378 67378 67378 67378 67378 67378 67378
Clusters 129 129 129 129 129 129 129
Adj. R2 0.530 0.532 0.530 0.530 0.531 0.531 0.531
Notes: Table A16 reports the coefficients every five years. The composition of banks is given by the log of
the total number of each type of bank, calculated every 5 years. Countries that did not have any of a type
of bank are given ln values of zero. Standard errors in brackets are clustered by the origin country. *p <
0.1, **p < 0.05, ***p < 0.01
157
A.2 Additional figures
Figure A1: Industry composition of global exports in 1865
textile fibers (26)

cereals (04)
bullion
textiles (65)
sugars (06)
coffee, tea, cocoa, spices (07)
beverages (11)
tobacco (12)
medicines (54)
animal & veg materials (29)
animal fats and oils (42)
dairy and birds' eggs (02)
hides (21)
non-ferrous metals (68)
miscellaneous (89)
meat and meat preparations (01)
cork & wood (24)
live animals other than fish (00
vegetables and fruit (05)
dyes (53)
fish (03)
coal (32)
crude minerals (27)
petroleum prod (33)
iron (67)
metal manu (69)
inorganic chemicals (52)
organic chemicals (51)
machinery (71)
paper (64)
metal ores (28)
nonmetallic mineral manu (66)
wood manu (63)
leather goods (61)
rubber (23)
soap (55)
misc edibles (08)
clothing (84)
0 20 40 60 80
Total global value of exports by SITC code in 1865 (m GBP)
Notes: Figure A1 shows the total value of world exports across all countries by two-digit SITC
categorization. The handcoded SITC category is given in parentheses next to the category name. Units are
millions of pounds sterling in 1865. Sources: Statistical Tables relating to Foreign Countries and Statistical
Tables relating to the Colonial and Other Possessions of the United Kingdom published in 1866.
158
Figure A2: Distribution of exposure to bank failure
(a) Ports (b) Countries
.6
.8
mean = 0.07
mean = 0.11
.6
.4
Density (fraction)
Density (fraction)
.4
.2
.2
0
0
0 .2 .4 .6 .8 1 0 .2 .4 .6
Port-level exposure to bank failure Country-level exposure to bank failure
Notes: Figure A2 plots the histogram of port (n = 289) and country (n = 55) exposure to bank failures for
the sample of ports and countries that were active in the pre-crisis year.
Figure A3: Positive correlation between country-level number of ships and exports values
150
Exports (millions GBP)
50 0 100
0 5 10 15
Ships (thousands)
1865 1866 1867 Fitted values

slope = 5.50, t-stat = 13.98
Notes: Figure A3 shows the positive linear relationship between the number of ships leaving a country in a
given calendar year (from the Lloyd’s List and the values of exports from that country. Three years around
the crisis year are plotted. The line is fitted to the pooled sample of all years.
159
Figure A4: Port-level effect of bank failures on exports: robustness to distance cutoffs
50
Cutoff distance from bank to port (km)

100
250
500
750
1000
1500
2000
-1.5 -1 -.5 0 .5
Confidence Interval
99% 95%
Notes: Figure A4 plots the estimated coefficients for b for the specification below, where the control group
of completely unexposed ports is based on the distance between the port and the nearest city of financing.
The baseline specification in the paper uses a cut-off of 500 km. ln(S pot ) = bFail po ⇥ Postt + a p + got + # pot
Figure A5: Positive correlation between sailing distance and geodesic distance
20
15
Sailing distance ('000 km)
10
0
0 5 10 15
Geodesic distance ('000 km)
Notes: Figure A5 plots the binscatter relationship between ports’ distance to each other measured
geodesically in kilometers and sailing distance measured in kilometers. The data for sailing distance come
from Philips’ Centenary Mercantile Marine Atlas II published in 1935. Sailing distances are calculated without
the Suez Canal route, which only opened in 1869. See appendix A.6 for a full discussion of the data source.
Geodesic distances are calculated based on the port’s longitude and latitude coordinates.
160
Figure A6: Positive correlation between news lag and geodesic distance to London
PAPEETE
100
AUCKLAND
80
YANTAI
News Lag (Days)
60
CAPE OF GOOD HOPE

40
RIO JANEIRO
20
BOSTON
BORDEAUX
0
0 5000 10000 15000 20000

Distance to London (Geodisic)
Notes: Figure A6 plots the relationship between the ports’ physical distance to London (measured
geodesically in kilometers) and the news lag in days that the ports received news of the banking crisis. The
circles convey the pre-crisis size of the port. Select ports from each continent are named.
161
Figure A7: Effect of above average exposure to bank failure on total exports
ln(EXot ) = b t I (Above avg exposureo ) + go + gt + # ot
.5
βt: treatment effect of above-average exposure
-.5
-1
-1.5
60
61
62
63
64
1 86 5
66
67
68
1 86 9
70
71
72
73
74
75
76
77
78
1 87 9
80
81
82
83
84
85
86
87
88
1 88 9
90
91
92
93
94
95
96
97
98
1 99 9
00
01
02
03
04
05
06
07
08
1 90 9
10
11
12
13
14
18
Years
Notes: Figure A7 plots b t from 1860–1914 for the specification above. The dependent variable is the ln of
the total value of exports for origin country o in year t. go and gt are country and year fixed effects,
respectively. The regressions are weighted by the total value of exports in order to most closely mirror
Figure 1.4. N = 5,799.
162
Figure A8: Annual country-level growth rates of total exports
.5
Exports: Annual Growth Rate
.25
0
-.25
1855 1865 1875 1885 1895 1905 1915

Year
Notes: Figure A8 plots the annual growth rates for the two groups of countries for the years before and
after the crisis. Calculated from the aggregate data presented in Figure 1.4. The vertical line marks 1866.
Figure A9: Aggregate GDP, grouped by above and below average exposure to bank failures
5
4
gdp (1866=1)
3
2
1
0
1845 1855 1865 1875 1885 1895 1905 1915

Year
Notes: Figure A9 plots the raw data for the total value of GDP by groups of countries, binned by above and
below average exposure to failure. GDP is normalized to equal 1 in 1866. The vertical line marks 1866.
163
Figure A10: Effect of bank failures from 1850–2014
1
-1
-2
-3
-4
-5
56 5
6 0
6 5
7 0
7 5
8 0
8 5
18 891 90
- 5
0 0
W 1 06- 5
I 1 9 10
6- 15
2 0
2 5
W 1 31 0
W 9 -35
4 0
4 5
5 0
5 5
6 0
6 5
7 0
7 5
8 0
8 5
19 991 90
- 5
01 0
0 5
11 0
4
18 185
18 -6
18 1-6
18 6-7
18 1-7
18 6-8
18 1-8
96 -9
19 190
19 1-0
19 192
19 1-2
19 6-3
19 -4
19 1-4
19 6-5
19 1-5
19 6-6
19 1-6
19 6-7
19 1-7
19 6-8
19 1-8
96 -9
20 200
20 -0
20 6-1
-1
1 6-
91 11-
1 6-
II 35
-
50
18
Years
Notes: Figure A10 plots the point estimates and 95 percent confidence intervals for the specification given
above estimated on the country-level panel of trade from 1850–2014. The dependent variable is the ln value
of exports. The specification includes origin country o FE, destination country-year dt FE, and time-varying
controls for the bilateral distance between countries. b t is the treatment coefficient on the effect of exposure
to failed banks on exports in each group of years. The coefficients most closely corresponding to WWI and
WWII are marked separately. Standard errors are clustered by the origin country. N = 665,866. Table A8
reports the coefficients.
164
Figure A11: Growth of multinational banks, 1850-1913
(a) 1850 (b) 1871
(c) 1891 (d) 1913
Notes: Figure A11 maps the total number of multinational banks (of all nationalities) from the Banking
Almanac for various years.
165
Figure A12: Banking sector recovery
(a) All banks (b) British banks

50
30
Average number of banks (1866=1)

40
20
20 30
10
10 0
0
1840 1860 1880 1900 1920 1840 1860 1880 1900 1920
year year
Above avg failure Below avg failure Above avg failure Below avg failure
(c) French banks (d) Domestic banks

150
80

60
100
40
50
20
0
1840 1860 1880 1900 1920 1840 1860 1880 1900 1920
year year
Above avg failure Below avg failure Above avg failure Below avg failure
Notes: Figure A12 plots the raw data of the average number of banks in cities exposed to above and below
average British bank failure. The data come from 5 year intervals of the Banking Almanac. Subfigure (a)
plots all the average for all banks. Subfigures (b), (c), and (d) split the total by nationalities. "Local banks"
refers to banks of the same nationality as the country it is located in. Each series is normalized to equal 1
in 1866. The vertical line marks 1866.
166
Figure A13: Effect of exposure to bank failure on new vs pre-existing trade relationships
βk(t) -2
-4
-6
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
1 86 5
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
1 99 9
00
01
02
03
04
05
06
07
08
09
10
11
12
13
14
18
Years
Failure x old relationship x year Failure x new relationship x year
Observations = 56830 , countries = 70, unweighted
Notes: Figure A13 plots the point estimates and 95 percent confidence intervals from the country-level
panel of trade in the specification given below. “Failure ⇥ old relationships x year" is the treatment
coefficient on the effect of exposure to failed banks on exports for bilateral trade relationships that existed
prior to 1866. “Failure ⇥ new relationships x year" is the treatment coefficient on the effect of exposure to
failed banks on exports for bilateral trade relationships that were newly formed after 1866. The dependent
variable is the ln value of exports. The specification includes origin country o FE, destination country-year
dt FE, time-varying controls for the bilateral distance between countries. Standard errors are clustered by
the origin country: ln(EXodt ) = b t,old Failo ⇥ I (Oldod ) + b t,new Failo ⇥ I (Newod ) + go + gdt + # odt
167
Figure A14: Treatment placebo
βt 0
-2
-4
1855 1860 1870 1875 1880 1885 1890 1895 1900 1905 1910 1915
Distribution of βt coefficients

Notes: Figure A14 plots the median, 25th and 75th percentile (edges of the box), and lower and upper
adjacent values for the frequency distribution of estimates of b t from running 1,000 regressions on
simulated data corresponding to equation 1.8 (above). The simulated data are generated from randomly
replacing the the country-level exposure to failure Failo with the exposure from another country. The end
year for each b’s range of year is given on the x-axis (for instance, 1855 refers to b 1850 1855 ).
Figure A15: Exports correlation within country regions

1
Fraction of exports in top 3 SITC groups
.8
.6
.4
.2
0
N. America Caribbean S. Africa N. Africa N. Europe S. Asia Oceania

C. America S. America W. Africa S. Europe E. Europe E. Asia
Notes: Figure A15 plots the fraction of exports in the top 3 SITC groups for each region. Exports values are
calculated from 1865. The full list of countries and their geographic regions are given in appendix A.5.4.
Regions are listed by geographic proximity, beginning in North America and traveling south and east.
168
Figure A16: Country region placebo
.15
.15
.15
.1
.1
.1
.05
.05
.05
0
0
-4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4
1850-55 1856-60 1866-70
.15
.15
.15
.1
.1
.1
.05
.05
.05
Density (fraction)
0
-4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4
1871-75 1876-80 1881-85
.15
.15
.15
.1
.1
.1
.05
.05
.05
0
-4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4
1886-90 1891-95 1896-1900
.15
.15
.15
.1
.1
.1
.05
.05
.05
0
-4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4
1901-05 1906-10 1911-14
Distribution of β coefficients
ln(EXodt ) = b t Failo + G0 Regionot + go + gdt + qt ln(dist)od + # odt

Notes: Figure A16 plots the frequency distribution of estimates of b t from running 1,000 regressions
corresponding to equation 1.8 (above) including origin-country region-year fixed effects, where the
origin-country is randomly assigned to a geographic region. The x-axis of each subfigure plots the
magnitude of the estimates for each group of years. The baseline impact of exposure to bank failures on
exports, estimated in column (8) of Appendix Table A7, is plotted as the thicker red dashed line, while the
mean placebo estimate (averaging across the 1,000 estimates) is plotted as the thin black dashed line.
169
Figure A17: Effect of other-country exposure within region on own country
2
θt
-2
-4
-55 856-60 61-65 66-70 71-75 76-80 81-85 86-90 91-95 96-00 901-05 06-10 11-14
1850 1 1
Years
Other country failure within region Own failure
Notes: Figure A17 plots the estimated coefficients from the regression specification below. "Own failure"
refers to the country-level exposure to failure Failo . "Other country failure within region" is the average
exposure to bank failure experienced by all other countries in the same geographic region. The dependent
variable is the ln value of exports. The specification includes origin country o FE, destination country-year
dt FE, time-varying controls for the bilateral distance between countries. Standard errors are clustered by
origin country: ln(EXodt ) = qt Failo ⇥ Failo,other + b t Failo + go + gdt + # odt
170
Figure A18: Recovery within country groups
(a) Within SITC groups
-1
βt
-2
-3
5 0 5 0 5 0 5 0 5 0 5 0 11-14
-185 56-6 61-6 66-7 71-7 76-8 81-8 86-9 91-9 96-0 1901-0 06-1
1850
Years
Observations = 48598 , countries = 48
(b) Within regions using the same countries
βt -1
-2
-3
5 0 5 0 5 0 5 0 5 0 5 0 11-14
-185 56-6 61-6 66-7 71-7 76-8 81-8 86-9 91-9 96-0 1901-0 06-1
1850
Years
Notes: Figure A18 plots the estimated coefficients from the regression specification below, which is the
main specification in equation 1.8 including SITC-year fixed effects (Figure A18a) and region-year fixed
effects (Figure A18b). Figure A18b is estimated on the same sample of countries as in Figure A18a, the
countries for which data on exports composition in1711865 is available.
A.3 Instrument validity
The empirical setting focuses on the effect of British credit contractions without observing
the share of British credit in total credit. This section shows that under certain assump-
tions, the instrument for British bank failures will recover the effects on credit contractions
from all banks.
Assume that the true model of the world is the following where Credittotal
l denotes
the total change in credit available in location l:
Dln(Yl ) = b 0 + d1 (DCredittotal
l ) + #l (A.1)
DCredittotal
l can be rewritten in terms of the share of total credit from British banks ab
and the share from non-British banks 1 ab :
DCredittotal
l = ab DCreditlBrit + (1 ab )DCreditnon
l
Brit
This allows us to rewrite Equation A.1 in the following way where b 1 = ab ⇤ d1 and
b 2 = (1 ab ) ⇤ d1 :
Dln(Yl ) = b 0 + b 1 DCreditlBrit + b 2 DCreditnon

l
Brit
+ #l (A.2)
Instrumenting for total credit loss using British bank failures:
DCredittotal
l = g0 + gB Faill,B + nl
# p ? 1, Faill,B and nl ? 1, Faill,B
The 2SLS estimator is:
Cov[Dln(Yl ), Faill,B ] b 1 Cov[DCreditlBrit , Faill,B ] + b 2 Cov[DCreditnon Brit

, Faill,B ]
d̂1IV ! = l
Cov[DCredittotal
l , Faill,B ] ab Cov[DCreditlBrit , Faill,B ] + (1 non Brit
ab )Cov[DCreditl , Faill,B ]
172
I use DCreditlBrit to proxy for DCredittotal
l .
1. d1 = b 1 when ab = 1, so b 2 = 0: non-British credit does not matter for trade
2. d̂1IV ! b 1 when: Cov[DCreditnon

l
Brit
, Faill,B ] = 0: the failure rates of British banks
in ports is not related to the change in non-British credit
173
A.4 Additional evidence on long-term effects
A.4.1 Lower-cost financing for shorter routes
Shorter routes are less expensive to finance because goods spend less time in transit. An
externally financed loan has shorter maturity, and it is easier for exporters to internally
finance out of working capital. Since financing costs increase with the distance between
trading partners, the key prediction is that trade between more distant partners will
decline after the bank failures.
I test this prediction using the panel of country-level values of trade by allowing
for the exposure to failure to differentially affect trading partners that are physically
closer. I construct a binary variable “Close” to indicate country-pairs that are less than the
average distance between countries trading in 1865. The results are robust to constructing
the variable over all years or at the end of the sample in 1914. Formally, I estimate the
following:
ln(EXodt ) = qt,close Failo ⇥ 1(Closeod ) + b t Failo + lt 1(Closeod )

(A.3)
0
+ Y Xod + go + gdt + # odt
Figure A19 plots qt,close in blue and b t in orange. b t , the effect of exposure to bank failure,
is very similar to the baseline effect in previous estimations. qt,close > 0 indicates that
conditional on exposure to bank failures, exports to closer destinations are positively
affected. The main effect for exports to close destinations is given by qt,close + b t , which is
close to zero. The qualitative interpretation is that a country’s exports losses are borne by
more distant trading partners, and that exporters are diverting their goods to destinations
with lower trade costs.
174
Figure A19: Exports are not affected for closer destinations
θt 0
-2
-4
-60 61-6
5
66-7
0
71-7
5
76-8
0
81-8
5
86-9
0
91-9
5 0
96-0 1901-0
5
06-1
0 11-14
1856
Years
Failure x Close Failure
Notes: Figure A19 plots the plots the qt and b t point estimates and 95% confidence intervals from the
country-level panel of trade in the specification given in equation A.3. The dependent variable is the ln
value of exports. The specification includes origin country o FE, destination country-year dt FE, and
time-varying indicators for common land border, common European colony, and common language.
“Failure ⇥ Close" is the treatment coefficient on the effect of exposure to failed banks on exports to
countries that are less than the average distance away from the destination country, where the average is
measured by 1865 bilateral trade flows. Standard errors are clustered by the origin country. N = 66,791.
175
A.5 Additional historical context
A.5.1 Trade finance
The mechanics of trade finance in the 19th century were conducted through bills of
exchange traded among the networks of banks and interbank lenders centered on Lon-
don. Bills were short-term loans that became contractual obligations when the creditor
“accepted” it by signing across it. In their simplest form, bills of exchange allowed for
debts between two parties. They were orders written by the “drawer” (lender) that the
“drawee” (borrower) would pay the face value of the bill (to the drawer, someone else, or
the bearer) at some point in the future. A check is simply a bill of exchange in the case
when the drawee is the drawer’s bank. A promissory note is a promise to pay between
the drawer and payee, where there is no drawee responsible for making the payment.
Bills usually had a maturity of 3-6 months (Cassis, 2016, p.93). The Treasury Bill was
proposed by Walter Bagehot in 1877 and modeled after the commercial bill of exchange
to allow the government to borrow at short maturities just as commercial interests were
able to.
British banks lent to their customers by “accepting” the customer’s bills of exchange.
British commercial law stipulated that the acceptor in turn became liable for the bill, such
that if the original borrower defaulted, the acceptor was responsible for payment. This
liability meant that acceptors transformed the idiosyncratic risk of individual borrowers
into their own credit risk. Bankers’ acceptances therefore bore the credit risk of the bank,
with the banks absorbing their customers’ credit risk. This guarantee made it easier to
re-sell the bills because the credit risk was easily observable. The acceptor would then
re-sell the bill to another individual or financial institution by “discounting” it on the
money market in London (Jones, 2000, p.23). The London money market’s liquidity came
from the size of the foreign bills market, and banks almost never held their own bills until
176
maturity (King, 1936). Discounts most resemble a modern-day repurchase agreement:
the seller received the face value minus the discount rate (haircut) at the initiation of
the transaction, and he paid the full face value in return for the security at its maturity.
At maturity, the bill was presented to the original borrower via his accepting bank for
repayment, and the debt terminated.
The term Discount Window in reference to the central bank comes from the fact
that bills of exchange were "discounted" there. Banks obtained emergency liquidity by
entering into a repurchase agreement with the Bank fo England on the short-term liquid
assets that it held. These assets were predominantly the bills of exchange that had been
extended by other banks abroad, reflecting the lending activity of those banks.
British multinational banks had accounts at the Bank of England, which in practice
meant that any security originated by one of these multinational banks was considered
high enough quality to be discounted at the Bank of England. The features of repo
agreements and joint liability protected the London money market from issues stemming
from asymmetric information where acceptors knowingly passed on bad bills. Those
features made their quality easily ascertainable, and bills were flexible and customizable,
so they became useful debt and investment instruments around the world, analogous to
commercial paper today. Although bills could be used for any purpose, those that origi-
nated outside of the United Kingdom primarily financed trade and were collateralized by
shipments.1
British multinational banks began being established in the 1830s both within and
beyond the British Empire to facilitate international capital flows, with the specific
purpose of increasing trade abroad. These banks were headquartered and raised capital
1 Cassis (2016) writes: “Finance required by the growth of international trade was supplied by private
bankers, increasingly by a small group of largely London-based merchant bankers who specialized in trade
credit by accepting bills of exchange and thus guaranteeing by their undoubted standing the payment of
the bills involved. The merchant banks’ backing was made clear by their acceptance on presentation of
the international trade bills with which they were individually connected. These providers of commercial
finance became known in the City as ‘acceptance houses’, and the paper involved as ‘acceptances’. The
bills were readily traded on the London market and so were liquid over the period, normally 60-90 days,
between their acceptance and maturity.” (p. 93)
177
in London by issuing deposits and shares, but they operated outside of Britain through
subsidiaries in cities around the world. The fact that they raised shares, issued deposits,
and invested abroad signaled a new movement in banking. These were the first “universal
banks” which then spread to Continental Europe in the subsequent decades (Cassis, 2016,
p.96). They most often funded the British merchants already established in foreign ports.
The lack of infrastructure in most countries was such that those merchants had to arrange
for their own financing and insurance if they wanted it. Their local knowledge was
invaluable to business, and the multinational bank subsidiary offices maintained close
contact with these exporters (Jones, 2000, p.27). See Table A17 for examples of these
banks and their operating regions.
Table A17: Examples of banks and operating regions
Bank Founding Year Operational Region
Anglo-Egyptian Bank 1864 Egypt, Mediterranean
Anglo-Italian Bank 1864 Italy, France
Bank of Australasia 1835 Australia, New Zealand
Bank of British North America 1836 Canada, USA
Chartered Mercantile Bank of India, Lon- 1853 India, China, Canada, Australia, Indone-
don & China sia, USA
Colonial Bank 1837 Caribbean
Imperial Bank 1862 Europe, Egypt, North America
Ionian Bank 1839 Greece
London Bank of Mexico & South America 1864 Mexico, Peru
Union Bank of London 1839 Australia, New Zealand, South America,

Asia, North America
Notes: This is a sample of the banks providing trade credit. The operational region is given as countries
although city-level variation is used in all the empirics. Sources: Bank of England Archives C24/1, Banker’s
Magazine, select bank histories listed in Appendix A.6.
178
A.5.2 Overend & Gurney
Transcript of the prospectus published on July 13, 1865
THE COMPANY is formed for the purpose of carrying into effect an arrangement which
has been made for the purchase from Messrs. Overend Gurney and Co., of their long
established business as bill brokers and money dealers, and of the premises in which the
business is conducted, the consideration for the goodwill being £500,000, one half being
paid in cash and the remainder in shares of the company with £15 per share credited
thereon – terms which, in the opinion of the directors, cannot fail to ensure a highly
remunerative return to the shareholders.
The business will be handed over to the new company on the 1st of August next,
the vendors guaranteeing the company against any loss on the assets and liabilities
transferred.
Three of the members of the present firm have consented to join the board of the
new company, in which they will also retain a large pecuniary interest. Two of them
(Mr. Henry Edmund Gurney and Mr. Robert Birkbeck) will also occupy the position of
managing directors and undertake the general conduct of the business.
The ordinary business of the company will, under this arrangement, be carried on as
heretofore, with the advantage of the co-operation of the board of directors, who also
propose to retain the valuable services of the existing staff of the present establishment.
The directors will give their zealous attention to the cultivation of business of a
first-class character only, it being their conviction that they will thus most effectually
promote the prosperity of the company and the permanent interests of the shareholders.
Copies of the company’s Memorandum and Articles of Association, as well as the Deed
of Covenant in relation to the transfer of the business, can be inspected at the offices of
the solicitors of the company.
LONDON, July 12, 1865.
179
A.5.3 London banking crisis
Previous scholarship
Banker’s Magazine wrote the following about Overend & Gurney’s share issuance:
The transformation of Overend, Gurney and Co.’s far famed discount estab-
lishment into a joint stock company, marks another era in the history of limited
liability...we may confidently anticipate that the position of the new company
will be relatively as high as the standing of the house to whose business it
succeeds.
Walter Bagehot’s account of Overend and Gurney’s demise in Lombard Street blames the
entirety of the failure on the directors:
In six years [from 1860-1866], the immensely rich partners lost all their own
wealth, sold the business to the company, and then lost a large part of the
company’s capital. And these losses were made in a manner so reckless and
so foolish that one would think a child who had lent money in the City of
London would have lent it better. (p. 19)
Anna Schwartz writes the following:
Overend, Gurney in earlier years had been a solid conservative partnership,

one of the pillars of the City. About 1860, a younger generation then in charge
of the business became less circumspect in its lending operations, accepting
equity interests for unrepayable loans extended to ironworks and shipping
companies. Losses led to a decision to incorporate with the possibility of
turning over a new leaf. The new company was launched in 1865 just after the
conclusion of the US Civil War, when there was every reason to anticipate a
strong revival of demand for British exports, but the new company did not
live long enough to benefit from it.[...] when on 10 May Overend, Gurney shut
down, the market was shaken. The next day panic broke loose. (p.273)
Bank of England response
In order to calm the London market, the Governor of the Bank of England appealed to
the Chancellor of the Exchequer to suspend the Banking Act of 1844. The Banking Act
of 1844 was the foundation of the gold standard in Britain and required that the Bank
180
of England’s currency supply was tied to the gold supply. This would allow the Bank
of England to accommodate the demands for liquidity by issuing currency beyond the
gold reserve at the Bank of England and effectively suspend the gold standard. The
government gave its permission, and this was sufficient to calm the markets so that the
gold standard remained in place. £5.6 million was lent to banks in just the first two days
of the crisis, collateralized on the short-term securities that reflected London’s lending
relationships. Although £5.6 million almost drained the Bank of England of its gold
reserves, it was small compared to the size of the banking sector, whose balance sheets
were almost £5 million each. Although the Bank of England was praised for averting a
deeper crisis, the size of the intervention was small relative to the size of the market, and
12% of banks failed.
The Overend & Gurney failure has been written about extensively by historians and
has been credited as the one that cemented the Bank of England’s role as Lender of Last
Resort. It was the event which led Walter Bagehot, the editor of The Economist at the time,
to argue that the monetary authority should, in times of crisis, discount bills of good
quality in the amount demanded to creditable borrowers (Bagehot, 1873). Domestically,
the 1866 banking crisis is attributed with causing the failure of over 200 firms. The shock
on manufacturing led to protests and riots that ultimately contributed to the passing of
the Reform Act of 1867, which greatly expanded the franchise. This was also known as
the Second Reform Act (the first was in 1832) and roughly doubled the franchise among
adult males in England and Wales.
First, worse quality banks could still approach the Bank of England Discount Window
for funds as long as they held bills that they could post as collateral. A bill’s riskiness
was determined by the bank that underwrote the debt, not the bank that brought in
the bill for discount. It is apparent from the ledgers that banks discounted the bills
originally accepted by other institutions, not themselves. This pattern is consistent with
the historical accounts that banks did not usually hold their own bills to maturity but
181
rather immediately discounted them on the London money market. Second, it is unlikely
that worse banks held lower quality bills because all banker’s acceptances of the same
maturity were discounted at the same market rate in normal times. Third, the average
rejection rate at the Bank of England did not change during the crisis, indicating the Bank
did not appear to change its policy during the crisis. These characteristics help to address
the main concern that worse banks would not have been able to obtain liquidity from the
Bank of England.
Figure A20: Bank of England Discount Window lending in 1866
4 12
3 9
millions GBP
% discount
2 6
1 3
0 0
Jan 01 Apr 02 Jul 02 Oct 01 Dec 31
Date
Total lent by Bank of England Discount rate (%)

Market rate (%)
Notes: Figure A20 shows the total amount of lending by the Bank of England at its Discount Window. The
red vertical line marks May 11, 1866. Sources: Bank of England Archives C24/1
Transcript of the Minutes of the Bank of England Court of Directors, Saturday May
12, 1866:2
A Court of Directors at the Bank on Saturday, the 12 May 1866
2 The
archived minutes are available at:
http://www.bankofengland.co.uk/archive/Documents/archivedocs/codm/18111911/codm041866111866b1.pdf
182
Present: Henry Lancelot Holland, Esquire Governor; Thomas Newman Hunt, Esquire
Deputy Governor [...]
The Governor laid before the Court the following correspondence:
Bank of England, 11 May 1866.
To: The Right Honourable, The Chancellor of the Exchequer, M. P.
Sir,
We consider it to be our duty to lay before the Government the facts relating
to the extraordinary demands for assistance which have been made upon
the Bank of England today in consequence to the failure of Messrs Overend
Gurney & Co. We have advanced to the Bankers, Bill Brokers and Merchants
in London during the day upwards of four million Sterling upon the Security
of the Government Stock and Bills of Exchange – an unprecedented sum to
lend in one day, and which,therefore, we suppose, would be sufficient to meet
all their requirements; although the proportion of this sum which may have
been sent to the Country must materially affect the question.
We commenced this morning with a Reserve of £5,727,000—which has been

drawn upon so largely that we cannot calculate upon having so much as
£3,000,000 —this evening, making a fair allowance for what may be remaining
at the Branches.
We have not refused any legitimate application for assistance, and, unless the
money taken from the Bank is entirely withdrawn from circulation, there is no
reason to suppose that this Reserve is insufficient.
We have honor to be, Sir, your obedient servants.
H.L. Holland, Governor and T.M. Newman Hunt, Deputy Governor.
The Chancellor of the Exchequer’s response:
Downing Street, 11 May 1866.

To: The Governor and the Deputy Governor of the Bank of England
183
Gentlemen,
We have the honour to acknowledge the receipt of your letter of this day to
the Chancellor of the Exchequer, in which you state the course of action at
the Bank of England under the circumstances of sudden anxiety which have
arisen since the stoppages of Messrs Overend Gurney & Company (Limited)
yesterday.
We learn with regret that the Bank reserve, which stood, so recently as last
night, at a sum of about five millions and three quarters, has been reduced in
a single day, by the liberal answer of the Bank to the demands of commerce
during the hours of business, and by its just anxiety to avert disaster, to little
more than one half of that amount, or sum (actual for London and estimated
for Branches) not greatly exceeding three millions.
The accounts and representations, which have reached Her Majesty’s Gov-
ernment during the day, exhibit the state of things in the City as one of
extraordinary distress and apprehension. Indeed deputations composed of
persons of the greatest weight and influence, and representing alike the private
and the Joint Stock Banks of London, have presented themselves in Downing
Street, and have urged with unanimity and with earnestness the necessity of
some intervention on the part of the State, to allay the anxiety which prevails,
and which appears to have amounted through great part of the day to absolute
panic.
There are some important points in which the present crisis differs from those
of 1847 and 1857. Those periods were periods of mercantile distress, but the
vital consideration of banking credit does not appear to have been involved
in them, as it is in the present crisis. Again, the course of affairs was then
comparatively slow and measured, whereas the shock has in this instance
arrived with intense rapidity and the opportunity for deliberation is narrowed
in proportion. Lastly, the Reserve of the Bank of England has suffered a
diminution without precedent relatively to the time in which it has been
brought about, and, in view especially of this circumstance, Her Majesty’s
Government cannot doubt that it is their duty to adopt without delay the
measures which seem to them best calculated to compose the public mind,
and to avert the calamities which may threaten trade and industry.
Of them, the Directors of the Bank of England, proceeding upon the prudent
rules of action by which their administration is usually governed, shall find
that, in order to meet the wants of legitimate commerce, it is requisite to extend
their discounts and advances upon approved securities so as to require issues
of Notes beyond the limit fixed by law, Her Majesty’s Government recommend
184
that this necessity should be met immediately upon its occurrence, and in that
event they will not fail to make application to Parliament for its sanction.
No such discount or advance, however, should be granted at a rate of interest

less than ten per cent, and Her Majesty’s Government reserve it to themselves
to recommend, if they should see fit, the imposition of a higher rate. After
deduction by the Bank of whatever it may consider to be fair charge for its
risk, influences and trouble, the profits of these advances will accrue to the
public.
We have the honor to be, Gentlemen, your obedient servants.
Russell Gladstone, Chancellor of the Exchequer
Resolved that the Governors be requested to inform the First Lord of the Treasury, and
the Chancellor of the Exchequer that the Court is prepared to act in conformity with the
letter addressed to them yesterday.
Resolved that the minimum rate of discount on Bills not having more than 95 days to
run, be raised from 9 to 10%.
A.5.4 Country characteristics
Table A18: Countries by region and British empire status
Country ISO code Region British Empire

Australia AUS OCEA 1
Austria-Hungary AUTHUN ESTEUR 0
Azores AZORES STHEUR 0
Belgium BEL NWEUR 0
Brazil BRA STHAM 0
British Guiana GUY STHAM 1
British West Indies GBRWINDIES CARIB 1
Canada CAN NORAM 1
Cape of Good Hope ZAF STHAFR 1
Ceylon LKA STHASI 1
Chile CHL STHAM 0
185
China CHN ESTASI 0
Colombia COL STHAM 0
Cuba CUB CARIB 0
Curacao ANT CARIB 0
Danish West Indies VIR CARIB 0
Denmark DNK SCANDI 0
Egypt EGY NORAFR 0
France FRA NWEUR 0
Germany DEU NWEUR 0
Gibraltar GIB STHEUR 1
Greece GRC STHEUR 0
Guatemala GTM CTRAM 0
Hong Kong HKG ESTASI 1
India - British Possessions GBRIND STHASI 1
Italy ITA STHEUR 0
Jamaica JAM CARIB 1
Japan JPN ESTASI 0
Java IDN STHASI 0
Malta MLT STHEUR 1
Mauritius MUS STHAFR 1
Mexico MEX CTRAM 0
Netherlands NLD NWEUR 0
New Zealand NZL OCEA 1
Norway Sweden SWENOR SCANDI 0
Panama PAN CTRAM 0
Persia IRN MIDEST 0
Peru PER STHAM 0
Philippines PHL STHASI 0
Poland POL ESTEUR 0
Portugal PRT STHEUR 0
Puerto Rico PRI CARIB 0
Romania ROU ESTEUR 0
Russia RUS ESTEUR 0
Siam THA STHASI 0
Sierra Leone SLE WSTAFR 1
Spain ESP STHEUR 0
St Helena SHN STHAFR 1
Straits Settlements STRAITS STHASI 1
Trinidad and Tobago TTO CARIB 1
Turkey OTTO MIDEST 0
USA USA NORAM 0
Uruguay URY STHAM 0
Venezuela VEN STHAM 0
186
A.6 Historical data sources
A.6.1 Data constructed
Bank characteristics
I gathered the banks’ 1865 and 1866 balance sheets and histories from annual reports
published in Banker’s Magazine, Banking Almanac and Directory, and The Economist. These
data include their age, capital (equity financing), leverage ratio, and reserve ratio. Publicly
traded banks did not consistently publish balance sheets until 1890, and even then only
half the private banks did so (Michie, 2016). Prior to that legislation, banks had complete
freedom over whether they publicly disseminated their balance sheets, so this information
is not available for all banks.
Port-level panel of trade

The source for the port-level is the daily publications of the Lloyd’s List newspaper.
Lloyd’s employed agents in ports around the world to gather information on international
shipping activity to send back to London. The primary consumers of this newspaper
were insurance agents, merchants, and family members of ship crews. The reporting in
Lloyd’s List is organized by port, based on the distance to London spiraling outwards.
Under each port, ships are listed individually with their name, their captain’s name,
type of ship, whether they arrived to the port or sailed from it, the destination of their
movements, and the date of the event. Coastal (i.e. domestic) trade was omitted from
the records for non-British ports. Lloyd’s also usually listed the date the intelligence was
sent, as there was often a lag between then and when it would have been received for
publication.
Processing the scans of the original prints required a labor-intensive combination of
OCR (Optical Character Recognition), python word processing, and manual data entry.
Almost 420,000 unique shipping events were processed. Lloyd’s List is very geographically
187
precise, so ports located within 10 kilometers of each other are aggregated into one port
unit. An example is that Cape of Good Hope is distinguished from Cape of Good Hope
Point, which are in the same bay. Ports that were aggregated into the same geographic
unit are matched to the same city for banking services.
Country-level panel of trade

The country-level panel of bilateral trade includes over 68,000 observations for 130
countries from 1850-1914. The sources are Pascali (2017), Dedinger and Girard (2017),
Fouquin and Hugot (2016), and Mitchener and Weidenmier (2008), along with the
Statistical Tables published by the United Kingdom and United States. Measures of
bilateral resistance between countries, such as common language, land border, and
common colonial background were taken from Fouquin and Hugot (2016). I recalculate
geodesic distance based on the center of the standardized pre-WWI country borders.
Measures of GDP and population from Fouquin and Hugot (2016) were also recalculated
to reflect those borders.
Industry composition of exports

I collected the composition of exports by country pre-crisis from the Statistical Tables
relating to Foreign Countries and Statistical Tables relating to the Colonial and Other Possessions
of the United Kingdom published in 1866. Values of exports by types of goods were
converted from various currencies into nominal pounds sterling as necessary. The
types of goods were manually standardized according to Standard International Trade
Classification (SITC) codes version 4. Appendix figure A1 lists the value of exports by
SITC category.
Sailing distances between ports

The sailing distance between ports is reported in nautical miles in the Philips’ Centenary
Mercantile Marine Atlas II published in 1935. Distances for different sailing routes are
188
given, but I exclude the Suez Canal route because it was not open until 1869. The routes
that are allowed include the Kiel Canal, Cape of Good Hope, Strait of Magellan, Cape
Horn, and Torres Strait.
City-level panel of banks

I gathered the names and city-level locations of all banks operating around the world
from 1850-1913 using the annual editions of the Banking Almanac. The data from 1861-1867
are annual; for the rest of the period I digitized almanacs at 5-year intervals. These records
make it possible to observe the operations of non-British banks throughout the entire
period. Nationalities are not given in the original source, so I assign bank nationalities
based on the locations of their headquarter offices (when known), the source of their
capital (usually given in their individual histories), and their names and areas of operation.
This dataset contains over 55,000 unique bank-location observations.
A.6.2 Data collected
Conflicts
I use Sarkees and Wayman (2010) from the Correlates of War project for data on
inter-state, intra-state, and extra-state conflicts from 1850–2014 to document conflicts
within the exporter-country and between country-pairs. For inter-state wars, I standardize
country borders to coincide with pre-WWI borders, the same way as in the panel of
trade data. Wars that occurred within one country’s borders (for instance, the Second
Italian War of Independence in which regions of Italy fought each other) are included as
a conflict for the exporting country, but is not included in the dyadic war variable because
the outcomes do not include own-country trade. Intra-state conflicts are recorded as a
war within the state where it is occurring (for instance the United States for the US Civil
War). Extra-state conflicts are recorded as a war for the official state and are not included
189
in the dyadic calculations of conflict.
In the pre-period balance checks in Table 2.3, I include all conflicts that occurred or
were ongoing in 1865 and 1866. There are 11 countries involved in inter-state conflicts,
3 in intra-state conflicts, and 2 in extra-state conflicts. These include the Paraguayan
War (Paraguay, Argentina, Brazil, Uruguay), Austro-Prussian War (Austria-Hungary,
Germany), Chincha Islands War (Spain, Peru, Chile), Second French intervention in
Mexico (France, Mexico), Third Italian War of Independence (Italy, Austria-Hungary),
Taiping Rebellion (China), Cretan Revolt (Ottoman Empire), United States Civil War
(USA), Polish Rebellion in Siberia (Russia), Bhutan War (United Kingdom).
Monetary standard
I gathered the data on the monetary standard of each country in 1866 using published
monetary histories or the wikipedia article for each country’s historical currency. In
cases, like in the British West Indies, when the official currency (pegged to the pound in
gold) circulated alongside unofficial currencies (like the Spanish pieces of eight in silver),
I categorized the country as being “bimetallic.” The results are not sensitive to being
categorized by the official currency (gold in this case).
190
B | Appendix to Chapter 2
B.1 Additional tables
Table B1: Alternative sample periods for first stage regressions
1(National Bank)
(1) (2)
1(pop<6k) 0.185** 0.274***
(0.0904) (0.0848)
D Pop(1870:1880) 0.000109*** 0.000116***
(0.0000330) (0.0000299)
State FE Y Y
Start Year 1873 1877
End Year 1883 1887
N 155 147
Notes: B1 presents first stage results with alternative sample periods. The first column shows national bank
entry likelihood between 1873 and 1883, and the second between 1877 and 1887. In both samples, town
population was below 6,000 in 1870, and between 4,000 and 8,000 in 1880. Additionally, there was no
national bank as of the start year (1873 and 1877, respectively).
191
Table B2: Alternative samples for first stage regressions
Panel A: National Bank Entries and Town Population (2,000 to 10,000)

1(National Bank)
(1) (2) (3) (4) (5) (6) (7)
1(pop<6k) 0.134 0.342*** 0.331*** 0.356*** 0.240***
(0.0831) (0.0795) (0.0708) (0.0736) (0.0669)
Population in 1880 0.0000365*** 0.0000476*** 0.000179*** 0.000137***
(0.00000979) (0.0000120) (0.0000161) (0.0000168)
Population in 1870 -0.000175*** -0.0000947***
(0.0000155) (0.0000171)
D Pop(1870:1880) 0.000146*** 0.000177*** 0.000117***
(0.0000132) (0.0000146) (0.0000160)
State FE Y Y
N 826 826 826 826 826 826 826
Panel B: National Bank Entries and Town Population (3,000 to 9,000)

1(National Bank)
(1) (2) (3) (4) (5) (6) (7)
1(pop<6k) 0.0571 0.229** 0.298*** 0.361*** 0.294***
(0.115) (0.108) (0.0861) (0.106) (0.0835)
Population in 1880 0.0000141 0.0000234 0.000147*** 0.000156***
(0.0000196) (0.0000271) (0.0000294) (0.0000309)
Population in 1870 -0.000174*** -0.000130***
(0.0000220) (0.0000254)
D Pop(1870:1880) 0.000135*** 0.000168*** 0.000136***
(0.0000193) (0.0000214) (0.0000246)
State FE Y Y
N 333 333 333 333 333 333 333
Panel C: Population Cutoff and Presence of National Banks

1(National Bank in 1875) 1(National Bank)
(1) (2) (3) (4)
1(pop<6k) -0.194* -0.186*** -0.0747 -0.0629
(0.105) (0.0662) (0.108) (0.0681)
Population in 1880 -0.0000903** -0.00000662
(0.0000449) (0.0000462)
Population in 1870 0.0000860*** 0.000000177
(0.0000264) (0.0000271)
D Pop(1870:1880) -0.0000867*** -0.00000126
(0.0000252) (0.0000260)
State FE Y Y Y Y
Adj. R2 0.311 0.314 0.243 0.245
N 341 341 341 341
Notes: Panel A and B of Table B2 presents first stage results with alternative population ranges around the
6,000 cutoff. Panel A uses the sample where 1880 population 2 [2, 000, 10, 000] and Panel B uses the sample
where 1880 population 2 [3, 000, 9, 000] Panel C expands the main sample (population in 1880
2 [4, 000, 8, 000] and population in 1870 < 6,000) by adding towns already had national banks as of 1875.
192
Table B3: Placebo: National Banks and local business (county-level)
Panel A: Percentage changes in share of doctors

OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) -0.0113 0.0147 0.0654 0.0872
(0.0323) (0.0329) (0.0899) (0.0918)
1(pop<6k) 0.0229 0.0290
(0.0338) (0.0330)
Mean of Dep. Var. -0.132 -0.132 -0.132 -0.132 -0.132 -0.132
Std. Dev. of Dep. Var. 0.154 0.154 0.154 0.154 0.154 0.154
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Panel B: Percentage changes in share of teachers
OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 0.0305 -0.0278 -0.0400 -0.0835
(0.0795) (0.0816) (0.217) (0.224)
1(pop<6k) -0.0140 -0.0278
(0.0832) (0.0820)
Mean of Dep. Var. -0.0513 -0.0513 -0.0513 -0.0513 -0.0513 -0.0513
Std. Dev. of Dep. Var. 0.417 0.417 0.417 0.417 0.417 0.417
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Notes: Table B3 presents results from the IV estimates as well as reduced form estimates. Dependent
variables in Panel A and B are percentage changes in shares of doctors and teachers among males above 21
years old from 1880 census to 1900 census, respectively. Regressions are weighted by share of town
population in the sample in 1880. * p<0.1, ** p<0.05, *** p<0.01.
193
Table B4: Robustness: National Banks and growth in manufacturing production
Panel A: Scaling production by male labor between 18-44 in age

OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 63.12 89.92* 375.2*** 509.7**
(45.77) (50.25) (145.4) (200.3)
1(pop<6k) 164.8*** 172.2***
(58.27) (58.34)
Mean of Dep. Var. 108.0 108.0 108.0 108.0 108.0 108.0
Std. Dev. of Dep. Var. 320.1 320.1 320.1 320.1 320.1 320.1
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
Panel B: Inflation-adjusted manufacturing production per capita
OLS RF IV OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) 52.66 254.9** 81.31* 421.4**
(35.57) (127.1) (48.69) (183.1)
1(pop<6k) 86.14** 142.4**
(41.86) (56.98)
Scaled by Males above 21yrs Y Y Y
Scaled by Males between 18-44yrs Y Y Y
Mean of Dep. Var. 151.0 151.0 151.0 201.5 201.5 201.5
Std. Dev. of Dep. Var. 252.9 252.9 252.9 318.1 318.1 318.1
N 148 148 148 148 148 148
KP F-stat 10.68 10.68
Notes: Table B5 presents results from OLS, reduced form, as well as IV estimates. Dependent variable is
changes in manufacturing production per capita between 1880 and 1890. Control variables include number
of state banks in a town as of 1876, as well as number of railroads in 1880. Regressions are weighted by
share of town population in the sample in 1880. * p<0.1, ** p<0.05, *** p<0.01.
194
Table B5: Placebo: National Banks and local manufacturing sector (1870-1880)
OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) -48.69 -32.18 -61.85 -51.20
(39.71) (41.69) (140.4) (148.9)
1(pop<6k) -18.97 -14.69
(47.49) (47.37)
Mean of Dep. Var. -36.30 -36.30 -36.30 -36.30 -36.30 -36.30
Std. Dev. of Dep. Var. 217.9 217.9 217.9 217.9 217.9 217.9
N 147 147 147 147 147 147
KP F-stat 8.686 8.215
changes in manufacturing production per capita between 1870 and 1880. Control variables include number
of state banks in a town as of 1876, as well as number of railroads in 1880. Regressions are weighted by
share of town population in the sample in 1870. * p<0.1, ** p<0.05, *** p<0.01.
Table B6: Placebo: National Banks and local innovation activity (1870-1880)
OLS RF IV
(1) (2) (3) (4) (5) (6)
1(National Bank) -0.781 -1.094** -0.502 -0.849
(0.489) (0.548) (1.325) (1.743)
1(pop<6k) -0.220 -0.287
(0.644) (0.661)
Mean of Dep. Var. 0.793 0.793 0.793 0.793 0.793 0.793
Std. Dev. of Dep. Var. 1.982 1.982 1.982 1.982 1.982 1.982
N 148 148 148 148 148 148
KP F-stat 15.74 10.68
percentage change in total number of patents between 1860s(1861-1870) and 1870s(1871-1880). Control
variables include number of state banks in a town as of 1876, as well as number of railroads in 1880.
Regressions are weighted by share of town population in the sample in 1870. * p<0.1, ** p<0.05, ***
p<0.01.
195
B.2 Additional figures
Figure B1: Bank note from the Pittsfield Bank in Massachusetts
(a) Bank note from the Pittsfield Bank in Massachusetts
(b) Page from Hodges’ New Bank Note Safe-Guard
Notes: Figure B1(A) gives an example of a private banknote that was printed in 1853. The note is for twenty
dollars, redeemable for specie at the Pittsfield bank. Figure B1(B) displays an page from the Hodges’ New
Bank Note Safe-Guard, first published in 1859. It is an example of one of the many publications dedicated in
helping merchants and brokers to detect counterfeit bank notes. It describes the physical appearance of
over 10,000 bank notes, “embracing every genuine note issued in the United States and Canada." The
description for the Pittsfield bank $20 note from Figure B1(A) is shown in the bottom row of the first
column and accurately describes the note.
196
Figure B2: Distribution of national banks in 1885
Notes: Figure B2 plots the location distribution of all national banks as of 1885. Each area represents a
county. The white areas did not have a national bank as of 1885, and the lighter to darker shades represent
1-3, 4-20, and 21 or more national banks within the county.
197
Figure B3: Timing of entry
(a) New National Banks in sample
20
Number of Entries by Group
10 5
0 15
<6000, after <6000, before >6000, after >6000, before
(b) All new National Banks between 1876 and 1885

10
Coefficients relative to 1880
-5 0 -10 5
1876 1878 1880 1882 1884

Year
Notes: Panel A of Figure B3 shows number of new national banks in our main sample before or after
census publication year, and whether population in 1880 census crossed the 6,000 threshold. Panel B shows
coefficients and standard error bars of coefficients on year dummy variables from the following regression:
NumberNewBankss,y = b y ⇤ 1(year = y) + gs ⇤ 1(state = s) + es,y .
198
B.3 Additional notes on the first stage
B.3.1 First stage regressions in alternative samples
We choose the 4,000 to 8,000 population range in order to obtain a sample of towns that
are largely comparable and hence less likely to be subject to omitted variable bias. The
first stage regressions are robust and more statistically significant with wider population
ranges. In Panel A and B of Table B2, we present first stage regression results for
1880 population between 2,000 and 10,000, as well as between 3,000 and 9,000 in 1880,
respectively. In both cases, having fewer than 6,000 population is strongly associated
with the likelihood of obtaining a national bank once we control for population levels or
changes.
In Panel C of Table B2, we expand our main sample to include all towns that already
had a national bank as of 1875. The first two columns suggest that when all of these
towns faced lower entry cost in the 1870s, smaller towns were less likely to obtain a
national bank. Column (3) and (4) show that our 1880 population cutoff is not correlated
with the existence of national bank in 1885. This is likely due to state bank conversions—
larger towns were more likely to have state banks and subsequently more national banks
converted from those state banks. This force operates in the opposite direction as the
instrument. Although state banks usually had much lower capital requirements than
$50,000, in reality they often operated with much larger capital (Knox, 1900), and the
conversion to national bank charters was minimally constrained by the capital requirement
defined in the National Banking Act. We therefore focus on the changes in access to
national banks by studying the towns with no national bank as of 1875.
199
B.3.2 Timing of entry
National banks were allowed to maintain the capital levels as of the time of their charter,
so one particular endogeneity concern is that banks could accurately forecast economic
and population growth, and would rush to obtain a charter before the 1880 census was
published. For example, a town with population below 6,000 in 1870 correctly anticipateed
that it would cross the population threshold in 1880 and established a bank in 1879, before
population in census was updated. This behavior of “racing the census” would bias the
OLS estimates upward because the bank entry would be correlated with the outcomes.
However, it would actually weaken the first stage and the reduced form relationship in
the instrument.
In addition, we empirically correlate the new national bank entries to the years around
the new census publication. First we compare new national bank entries right before the
1880 census in our sample of towns, and then we look at the towns in the entire country.
Figure B3 presents the results. In Panel A, we compare number of new national bank
entries by population in 1880 census for the towns in our sample. The first and third
bars shows the number of new entries right after the new census was published, and the
second and fourth bars shows the number of new entries right before it. For towns below
the threshold, there is more entry, which is consistent with our first stage results. For
both population groups, more entry occurred after the new census was published rather
than before.
In Panel B, we look at national bank entry in the entire country. We present coefficients
on year dummy variables from the following equation:
NumberNewBankss,y = b y ⇤ 1(year = y) + gs ⇤ 1(state = s) + es,y , (B.1)
and compare the coefficients relative to 1880. The result also show that there was no spike
in entry before the new census and that in fact, more new national banks established
200
after 1881 than before. The empirical evidence suggests that there was no census racing
behavior.
In fact, our first stage results are robust to selecting any year ranges that starts prior
to the 1880 census and ends in the 1880s. We choose the range from 1875 to 1885 to
capture a relatively stable time period in terms of economic activities and relative growth
in national and state banks.
B.4 Additional historical context
B.4.1 Banking system before the National Banking Act
The idea of establishing a unified banking system across the United States was several
decades earlier than the passage of the National Banking Act. The First Bank of the
United States, charted for a term of twenty years by the Congress on February, 1791,
operated in Philadelphia and was the nation’s de facto central bank. Alexander Hamilton,
the first Secretary of the Treasury, believed a national bank was necessary to stabilize and
improve the nation’s credit, and proposed federal mint as common currency. However,
the bank faced widespread resistance due to concerns of expanding federal power, which
was famously led by the Secretary of State Thomas Jefferson. The bank charter was not
renewed and expired in 1811.
In 1816, the Second National Bank started operation with similar functions as the
First Bank of the United States. As of 1832, the Second National Bank operated more
than 30 branches nationwide. However, the political clashes over the power of a national
bank system eventually led to failure of its charter renewal in 1836, which marked the
beginning of the Free Banking Era.
201
B.4.2 Additional evidence of bank debt illiquidity in the Free Banking
Era
The large number of floating exchange rates created inconvenience in economic activity.
For example, a case record compilation of the United States supreme court between 1843
and 1846 (Stephen K. Williams, 1901) contains a case regarding the value of a loan, and
how its value had changed over time:
[...] the defendant did [...] receive the amount of said loans from the plaintiffs
in the bank notes of Virginia and of other States, which, [...] were depreciated
considerably below the current value of the bank notes of this district [...]
The frictions stemmed from state bank notes illiquidity was especially detrimental to
interstate transactions. As a contemporary traveler illustrated the magnitude of the cost
in his diary (Dewey, 1910):
Started from Virginia with Virginia money; reached the Ohio River; exchanged
$20 Virginia note for shinplasters and a $3 note of the Bank of West Union
[...] At Maysville wanted Virginia money; couldn’t get it.
[...] reached Fredericktown; there neither Virginia nor Kentucky money current;
paid a $5 Wheeling note for breakfast and dinner; received in change two
$1 notes of some Pennsylvania bank, $1 Baltimore and Ohio Railroad, and
balance in Good Intent shinplasters; 100 yards from the tavern door all notes
refused. [...]
202

Xu Dissertation 2019

Uploaded by

Copyright:

Available Formats

Xu Dissertation 2019

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Xu Dissertation 2019

Uploaded by

Copyright:

Available Formats

Essays on Finance in History

Share Your Story

The Department of Economics

in partial fulfillment of the requirements

All rights reserved.

Essays on Finance in History

1 Reshaping Global Trade: The Immediate and Long-Run Effects of Bank

2 National Banks and the Liabilities Channel of Local Economic Activity 63

3 Safety in the Gold Standard Era 116

A Appendix to Chapter 1 142

B Appendix to Chapter 2 191

1.1 Pre-crisis comparison of bank characteristics . . . . . . . . . . . . . . . . . . 21

2.1 Comparison of National Banks and state banks . . . . . . . . . . . . . . . . 76

A1 Bank-level relationship between failure and credit supply . . . . . . . . . . 142

B1 Alternative sample periods for first stage regressions . . . . . . . . . . . . . 191

1.1 Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.1 Relative discounts on state bank notes relative to Philadelphia . . . . . . . . 73

3.1 Banking Department balance sheet . . . . . . . . . . . . . . . . . . . . . . . . 120

A1 Industry composition of global exports in 1865 . . . . . . . . . . . . . . . . . 158

B1 Bank note from the Pittsfield Bank in Massachusetts . . . . . . . . . . . . . 196

This section provides an overview of the institutional structure of British multinational

1.2.1 Trade finance & British banking dominance

Contractual frictions were a major barrier to establishing international trading relation-

1.2.2 London banking crisis of 1866

1.3 Measuring exposure to London’s crisis around the world

The goal of my empirical analysis is to estimate the causal relationship between a

Faill = Â zlb,pre ⇥ I(Failureb ) (1.3)

D ln(Creditlt ) = a1 + b 1 Faill + G10 Xlt + nlt (1.4)

Faill is a location-level analogue to the firm-level exposures to bank-level shocks used

D ln( EXlt ) = a2 + b 2 Faillt + G20 Xlt + elt (1.5)

The reduced form coefficient b 2 in Equation 1.5 is straightforward to interpret as the

1.3.2 Validity of reduced form estimation

Panel A: Balance sheet characteristics (joint-stock banks)

All Not Failed Failed Diff

Panel B: Other characteristics (all banks)

All Not Failed Failed Diff

N 128 106 22 128

Correlation between location characteristics and bank failure rates

X̄b = a + bI (Failureb ) + # b (1.6)

some of them to reject the null.

Panel B: Country characteristics

Panel C: Country characteristics: exports composition

1.4.1 Exposure to bank failures

1.4.2 Immediate outcomes

(a) Excerpt of the Bank of England Discount Office ledgers

(b) Excerpt of the Lloyd’s List

(a) British multinational bank lending and failures

(b) Port-level trade activity

1.4.3 Long-term outcomes

1.5.1 Intensive margin effect: baseline specification

ln(S pot ) = bFail po ⇥ Postt + a p + got + # pot (1.7)

(a) Port-level relationship

(b) Country-level relationship

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

Allowing for news lags

estimations. These singleton observations account for 5 of the 2,532 observations.

Intensive margin effects: values of trade

ln(EXodt ) = b t Failo + G0 Xot + go + gdt + qt ln(dist)od + # odt (1.8)

1.5.2 Extensive margin effects

ln(EXodt ) = b t Failo + G0 Xot + go + gdt + qt ln(dist)od + # odt