Golden Fetters or Credit Boom Gone Bust?
A Reassessment of Capital Flows in the Interwar Period*
Lukas Diebold†
April 2025
Abstract
This paper uses newly digitized Balance of Payments data for 33 countries to study international capital flows and their economic implications during the interwar period. I document
the boom-bust pattern in capital flows centered on the Great Depression and show that gross
foreign credit is the decisive link between capital flows, reduced output growth, financial crises,
and post-crisis recession severity. The Gold Standard played a key role by exposing countries to
foreign capital through global financial integration, while simultaneously restricting their ability
to respond to surging inflows. Using an instrumental variable based on bilateral exposure to the
US, foreign capital supply shocks are identified as an important driver of these dynamics.
Keywords: Capital flows, business cycles, financial crises, economic history
JEL classification codes: F34, G01, G15, N10
* I am thankful to Philipp Ager, Edoardo Acabbi, Ivan Alfaro, Alexander Donges, Rui Pedro Esteves, Lukas Hack, Max
Jager, David Koll, Dmitry Kuvshinov, David Martinez-Miera, Christopher Meissner, Kris Mitchener, Eric Monnet, Gonçalo
Pina, Björn Richter, Albrecht Ritschl, Gianmarco Ruzzier, Jochen Streb, Kirsten Wandschneider, Robert Wojciechowski,
attendees at the V. Congress for Economic and Social History, the EHS Annual Conference 2023, the 2023 ifo-workshop on
Macroeconomics and International Finance, the 15th Conference of the European Historical Economics Society, the 9th
joint CEPR and Banco de Espana Economic History Seminar, the London School of Economics Graduate Economic History
Seminar, the 2024 meeting of the Spanish finance association, the CEPR’s International Macro History Online Seminar
Series, the Society for Economic Dynamics’ Winter Meeting, as well as seminar participants at Universidad Carlos III de
Madrid, Cunef, Luiss Guido Carli, Corvinus University Budapest, the University of Tübingen and Mannheim University
for helpful comments and suggestions. I am also thankful to Dennis Quinn for sharing Data and Nico Spiegel and Antea
Kurtovic for excellent research assistance. Financial support from the Joachim Herz Foundation, and the Young Researchers
Fund of UC3M are gratefully acknowledged.
† Lukas Diebold: University Carlos III de Madrid; ldiebold[at]emp[dot]uc3m[dot]es
1.
Introduction
In 1929, few believed that the world economy was on the brink of its first truly global economic crisis
(Irwin, 2014), and more than 60 years later Bernanke (1994) stated that we are still a long way off
from understanding ’the holy grail of macroeconomics’: the Great Depression. We do know, however,
that the unique duration and severity of the Great Depression was linked to fragile financial systems
and capital markets (Bernanke and James, 1990; Schnabel, 2004; Bernanke, 2009). But where did
this fragility come from? The international nature of the depression already led Fisher (1935) and
Keynes (1941) to think about the role of the global financial system in creating and transmitting
financial fragility. Kindleberger (1978) later added that foreign financing amplified the boom-bust
pattern around crises.1 Crucially, their approaches center on gold or net capital flows, like the current
account, but we still know very little about gross international capital flows and their implications
during the interwar era (Accominotti and Eichengreen, 2016). Gross flows, however, not only dwarf
net flows in size but also better capture the buildup of external debt and exposure to global capital
movements (Shin, 2012; Broner et al., 2013). Honing in on them, consequently, allows for a better
understanding of the relationship between capital flows, financial cycles, and economic crises in
general, while offering a new angle on the Great Depression in particular.
In this paper, I go beyond the traditional emphasis on gold and net capital flows and, using newly
digitized data for 33 countries between 1923 and 1938, study the dynamics of gross capital flows
and their implications for real economic outcomes in the interwar period. My results show that
exposure to gross foreign credit is the most important predictor of business cycle downturns, financial
crises, and recession severity. Net flows, in contrast, capture this exposure only imperfectly and are
insignificant whenever gross flows are included in the analysis. Recently, variation in the global
supply of capital has been identified as an important driver of domestic financial conditions (Rey,
2013; Miranda-Agrippino and Rey, 2020). Using the interaction of bilateral exposure to US-foreign
investments with the total change in the US-foreign investment position as an instrument, I show
that foreign capital supply is also crucial to understanding capital flow dynamics in the interwar
period. In doing so, I offer an alternative interpretation of the influential ’Golden Fetters’ thesis
(Eichengreen, 1992), and argue that the Gold Standard created exposure to gross foreign inflows by
integrating countries into the global financial system, while at the same time restricting the scope of
action for governments to respond to surging capital flows.
Central to this analysis are the Balance of Payments (BoP) statistics from the League of Nations
(1930-1932, 1933-1939a), compiled into a novel dataset and reevaluated to address concerns about
data quality raised by previous authors. I begin by demonstrating that the new data accurately
captures gold movements, and the international capital (im-) balances revolving around German
1 The idea to explain domestic conditions with international finance is older still, as already Hume (1758) regarded the
management of the external balance as vital for the supply of gold and domestic stability.
1
reparations, known from previous literature.2 Building on this, I then leverage the newly collected
series on gross capital flows, revealing that gold (its role as a monetary multiplier during the Gold
Standard notwithstanding) and net capital flows only reflect a fraction of capital movements in the
interwar period. Notably, the ratio of net to gross flows reached a trough in 1929, hinting at the
importance of gross capital flows in characterizing the global financial system on the eve of the Great
Depression. Unlike gold and net flows, which lack the boom-bust pattern often tied to financial
cycles and crises (Reinhart and Rogoff, 2009; Schularick and Taylor, 2012; Jordà et al., 2013), gross
flows also display a pronounced boom until 1929, followed by a long-lasting bust.
To link capital flows to the interwar business cycle empirically, I first use local projections (Jordà,
2005), showing that gross inflows are followed by growth slowdowns in the medium term. The
results are economically meaningful, with a one standard deviation increase in gross inflows being
associated with lower cumulative growth of about 4 percentage points after 4 to 5 years. As capital
flows accumulate into international investment positions, I compute cumulative capital flows over
a three-year window and use them in predictive regressions of GDP, similar to Mian et al. (2017).
Again, gross inflows emerge as the single most significant predictor of economic downturns. The
responses in this alternative specification are larger than in the local projections, suggesting that the
effect of continued foreign inflows is, to some extent, additive. Both findings hold in a battery of
robustness checks, where neither net inflows, nor gross outflows, show comparable dynamics.
The interwar business cycle cannot be discussed without the crisis sitting at its heart. The
literature has shown that crises tend to be preceded by surges in international capital flows (Reinhart
and Rogoff, 2009; Caballero, 2016) and succeeded by capital flight and sudden stops in lending
(Broner et al., 2013; Krishnamurthy and Muir, 2017; Romer and Romer, 2017; Diebold and Richter,
2021). The Great Depression fits this pattern perfectly. Using a probit estimation, I confirm that
gross foreign inflows are the single most reliable predictor of financial crises. In fact, adding other
capital flow variables to a model already containing gross foreign inflows does not increase predictive
accuracy. Given the cyclicality of capital flows around the Great Depression these results align well
with economic intuition. The question is: can heterogeneous exposure to these dynamics before
crises explain the heterogeneity in real economic outcomes afterwards?
In a setting similar to Jordà et al. (2013), where the authors find that recessions become more
severe when the preceding domestic credit boom was large, I study recession severity conditional on
previous exposure to gross foreign inflows, and find that higher exposure amplifies crises. Again,
the result holds in a variety of robustness checks, including different measures of exposure. This
corresponds to theory developed in Caballero and Simsek (2020), where the fickleness of foreign
capital turns out to be harmful due to its run-like tendencies during crises. But where does the
2 The latter, also called ’debt carousel,’ refers to international lending after WWI driven by war debts and reparations.
See: Spoerer and Streb (2013) for a sketch of how the ’carousel’ was supposed to function, De Broeck et al. (2018) for an
estimation of bilateral flows, and End et al. (2019) for a detailed description of the involved financial instruments. For a
discussion of gold movements in the interwar period, see: James (1992) and Irwin (2012).
2
capital run to? Caballero and Simsek (2020) also argue that the repatriation of capital can moderate
recessions, as returning foreign assets can be used to buffer the effects of decreasing foreign capital
availability. Inverting the previous setting to analyze capital exports (instead of imports) before crisis,
I find empirical evidence for this channel. The accumulation of foreign assets helps to moderate
recessions, but cannot fully offset the negative implications of foreign liabilities.
The dynamics of gross capital flows help rationalize these results. By estimating a local projection
of capital flows on a financial crisis indicator, I show that while total capital flow volumes decrease
after crises, inflows and outflows respond differently. Outflows increase slightly (flight), while inflows
decline sharply (stops). Intuitively, the decrease in available capital is more severe for countries highly
reliant on inflows, whereas countries able to repatriate foreign assets are less affected. Combining
intuition with the framework from Forbes and Warnock (2012) to identify discrete capital flow
episodes (stops, surges, flight, retrenchment), I demonstrate more generally that high levels of
inflows combined with rapid reductions correlate with the worst growth trajectories and highest
crisis likelihood in subsequent years. Similarly, surges in inflows are followed by lower growth when
inflows are already high, but, interestingly, higher-than-average growth when levels are low.
Which mechanisms explain the link between gross foreign inflows and adverse outcomes? A key
factor in the surging capital flows of the interwar years was the Gold Standard. As the dominant
monetary system of the 1920s, it facilitated global financial integration, reduced currency risk, and
signaled a commitment to free capital movement (Bordo and Kydland, 2005; Wandschneider, 2008).
Consistently, I find that Gold Standard adoption was followed by a significant rise in gross foreign
inflows. At the same time, the Gold Standard’s ’fetters’ limited governments’ ability to respond to
increased foreign exposure via capital account management, exchange rate adjustments, or monetary
intervention during crises (Eichengreen, 1992). I find that countries off the Gold Standard, having
this margin of adjustment, were indeed able to partially insulate themselves from the adverse effects
of foreign credit. A similar result emerges when using a closed capital account measure, aligning
with Mitchener and Wandschneider (2015), who note that after leaving the Gold Standard, countries
introduced capital controls while the option for monetary policy remained underutilized. Capital
flows may also be influenced by adopting or leaving the Gold Standard due to the accompanying
shifts between official reserves (gold and foreign assets).3 When further investigating this relationship,
I find capital inflows (outflows) to be linked to lower (higher) official reserves, and especially foreign
asset reserves to be positively associated with GDP growth and less severe recessions.
Foreign inflows contribute to the expansion of domestic bank balance sheets by increasing
available capital within a country’s financial system (Magud et al., 2014). This heightened capital
availability, however, potentially amplifies the vulnerability of banking systems to capital flight and
sudden stops, as foreign capital typically carries shorter maturities than the domestic investments
3 In fact, changes in official reserves and private capital flows are often considered separate items (Broner et al., 2013).
3
it funds, undermining systemic stability (Obstfeld, 2012; Collet and Postel-Vinay, 2021). When
short-term funding dries up, financial institutions are often forced to liquidate illiquid (long-term)
assets to service the withdrawal of liquid (short-term) liabilities. This contractionary pressure on
bank balance sheets reduces credit creation, with adverse implications for future economic outcomes
(Adrian and Shin, 2010; Chodorow-Reich, 2014). Using newly digitized bank balance sheet data, I link
gross inflows to bank balance sheet expansions, which are characterized by short-term instruments
on the liability side and long-term instruments on the asset side, creating a mismatch in maturities.
Importantly, no comparable relationship exists between bank balance sheets and net inflows.
Borrowing today implies future interest payments to creditors (Drehmann et al., 2023). When the
recipients of these payments exhibit a lower marginal propensity to consume their income domestically, and debtors face financial constraints, this transfer depresses domestic demand (Eggertsson
and Krugman, 2012). As foreign creditors likely spend their incomes abroad, especially foreign credit
has long been acknowledged as a drain on aggregate domestic income (Lerner, 1948). Relying again
on data from the BoP, which reports ”interest and dividend payments to foreigners” as a current
account item, I confirm in the first step that this variable increases with past foreign inflows. In
the second step, I link contemporary interest payments to future GDP growth, finding that higher
present-time interest payments to foreigners are followed by reduced output growth.
A potential amplifier for both the balance sheet and debt service channel is the high cost associated
with defaulting on foreign debt (Reinhart and Rogoff, 2009). Correspondingly, Tomz and Wright
(2007) find only a weak relationship between economic downturns and foreign defaults, and argue
that the norm is to continue debt service in the face of adverse shocks.4 In practice, this may lead
countries to cut back on domestic spending first and foreign credit obligations second, further
exacerbating the reduction in domestically available capital and spending.
What drives the inflow of foreign capital into countries? On one hand, inflows may be determined
by idiosyncratic domestic demand for capital; on the other, by the supply of capital on international
markets, with my baseline being unable to distinguish between the two factors. Utilizing additional
data on bilateral portfolio investments of the United States, I isolate foreign capital supply by
constructing an instrument that interacts past bilateral exposure to the US with the present change in
the total US portfolio position (similar to Bartik (1991)). In addition to exploiting exogenous variation
in capital inflows, this also addresses a potential downward bias in the baseline, as borrowing abroad
against future fundamentals is unlikely to have adverse aggregate effects, while foreign supply,
unrelated to domestic conditions, is crucial for adverse outcomes (Rey, 2013; Miranda-Agrippino and
Rey, 2020). Consistent with this intuition, the instrumented coefficients of gross foreign inflows are
highly significant across specifications and increase relative to the OLS baseline.
4 Reinhart and Rogoff (2009) argue that this is due to foreign defaults cutting countries off from international markets for
extended periods and induce high penalties for future borrowing. Erce and Mallucci (2018) find that especially countries
with small domestic credit markets, reliant on foreign funding, selectively default on domestic rather than foreign debt.
4
The paper contributes to several strands of literature. First is the long history of the study of
capital flows. From Hume (1758) thinking about external balances, through Fisher (1935) asking if
capital flows transmit economic conditions, to long-run studies of external imbalances (Jordà et al.,
2011), the current account has been at the center of attention, but findings have largely depended
on sample composition.5 This first prompted the question: ’Does the current account still matter?’
(Edwards, 2002; Obstfeld, 2012), and ultimately the insight that ’we have asked the current account
to do too much’ (Borio, 2016). Recent literature has consequently argued for a greater focus on gross
measures (Shin, 2012; Calderon and Kubota, 2012; Borio and Disyatat, 2015), to which I contribute by
constructing a new dataset on gross capital movements, extending their documentation and analysis
into the interwar period. Justifiable concerns about the BoP’s data quality have previously prevented
this type of study (James, 1992; Feinstein and Watson, 1995). A thorough reassessment, however,
reveals that we may have underestimated the data’s potential insights, despite its limitations.
Second, the paper explores the relation between capital flows and economic outcomes. Surges
in capital flows have been shown to precede downturns in the business cycle and crises (Reinhart
and Rogoff, 2009; Caballero, 2016), followed by capital flight and contracting flow volumes (Forbes
and Warnock, 2012; Broner et al., 2013; Caballero and Simsek, 2020). This tends to amplify the
boom-bust-pattern of the business cycle (Kindleberger, 1978). While Kiley (2021) finds a link from
current account deficits to crises, other studies have shown that historically, crises are equally likely
in surplus and deficit countries (Obstfeld et al., 2010; Jordà et al., 2011). I show that any link between
net flows and the business cycle disappears whenever gross flows are included in the model. Gross
foreign inflows instead are robustly related to adverse outcomes across all specifications.
An important aspect in the relationship between capital flows and economic outcomes is the
dynamics associated with the international supply of capital. Global supply, independent of individual domestic economic conditions (Rey, 2013), has been shown to increase financial fragility due to
heightened risk of capital retrenchment (Milesi-Ferretti and Tille, 2011), run-like dynamics (Forbes
and Warnock, 2012; Caballero and Simsek, 2020), or ’sudden stops’ during crises (Calvo, 1998; Accominotti and Eichengreen, 2016). This paper confirms that countries able to shield themselves from
these dynamics are less affected by their consequences. Using a novel IV approach, I further show
that it is indeed supply-driven inflows that are responsible for the negative aggregate link between
capital inflows and economic outcomes. Gross flows are crucial in this context, as net flows neither
fully capture the supply of capital from international markets (net credit may fall while gross credit
expands), nor can all countries experience net inflows or net capital flight simultaneously. However, I
show that it is both plausible and consistent with empirical evidence that most countries experienced
expanding gross inflows before, and contracting gross inflows after, the Great Depression.
Naturally, the paper also aims to contribute to the literature on the interwar business cycle and the
5 See Adalet and Eichengreen (2007); Hoffmann and Woitek (2010); Jordà et al. (2011) for long-run and historical samples,
and Liadze et al. (2010); Mian et al. (2017); Kiley (2021) for more recent sample compositions.
5
Great Depression. The iconic boom-bust pattern around the Great Depression has been explained by
the systematic vulnerability and pro-cyclicality of financial systems (Bernanke and James, 1990), but
what makes financial systems vulnerable in the first place? The idea that it was the mismanagement
of a restrictive financial system, the Gold Standard, is manifested in the metaphor of the ’Golden
Fetters,’ which needed to be shed to break the downward spiral of the depression (Eichengreen, 1992;
Eichengreen and Temin, 2000; Ellison et al., 2023). This explanation is compelling, as it identifies a
common factor among a large sample of countries experiencing severe recessions. But while much of
the previous focus has been on what the Gold Standard hindered countries from doing domestically,
I focus on what it exposed countries to internationally.
In Eichengreen and Mitchener (2003), the authors narrate the Great Depression as a domestic
credit boom gone wrong, while Borio et al. (2014) argue that countries with large credit booms
prior to the Great Depression were connected via gross capital flows. Ritschl (1999) called for a
broader approach to interwar capital flows beyond the much studied role of US foreign lending and
Keynes (1941), drawing on his interwar experience, already linked contractionary biases in countries
heavily reliant on foreign capital to capital retrenchment, focusing on net measures. Additionally,
Accominotti and Eichengreen (2016) find a sudden stop in capital flows during the Depression and
Quinn (2003) shows that countries with more open capital accounts prior to the Great Depression
experienced deeper recessions. I combine these perspectives to tell the story of the Great Depression
as an international credit boom that went bust, more strongly affecting countries with previously high
levels of foreign inflows. For this, the Gold Standard’s international dimension was instrumental, as
it increased exposure to capital movements via integration into the global financial system, while
domestically, the Golden Fetters tied the hands of governments responding to the bust.
2.
Data and Balance of Payments Mechanics
My main data source are the Balance of Payments statistics, compiled by the League of Nations, first
published in three volumes from 1930 to 1932 under the title ’Memorandum on International Trade
and the Balance of Payments’ (League of Nations, 1930-1932). These first attempts at homogenized
national accounting include over 40 countries and cover the years 1922-1930. The format was
replaced in 1933 with the updated and revised ’Balance of Payments’ (League of Nations, 1933-1939a),
published in seven volumes starting in 1933 and covering the period 1929-1939. After digitizing all
volumes and dropping countries with less than five years of coverage, I obtain an unbalanced panel
of 33 countries. Since series frequently overlap across publications, more recent entries are used
first and extended backwards with earlier data. An example of the original publication is shown in
Figure A1.1. Initially, the data is collected in domestic currency, but each publication includes the
main aggregates for each country in US dollars, using the pre-1933 Gold-Dollar parity. From this, I
compute the yearly exchange rates used by the LoN and convert all data into US dollars.
6
Figure 1: Capital account composition for the U.S., Germany, the U.K., and Japan
In Billion USD$
-.25
0
.25
2
-.5
In Billion USD$
-1
0
1
-2
In Billion USD$
-1
0
1
1924 1927 1930 1933 1936
1924 1927 1930 1933 1936
Japan
.5
United Kingdom
2
Germany
-2
-4
In Billion USD$
-2
0
2
4
United States
Long
Short
Balance
1924 1927 1930 1933 1936
1924 1927 1930 1933 1936
Notes: This figure shows the annual gross financial flows from the capital account side of the Balance of Payments for the United States,
Germany, the United Kingdom and Japan. Figures are in billion US-dollars (note that scales are country specific). Capital inflows (credit)
are plotted as positive values, capital outflows (debit) as negative values. Blue and red represent flows in long- and short-term capital
flows respectively. The black line represents the capital account balance, with a positive balance indicating the net inflow of capital.
To address potential concerns about the data’s quality and discuss some of its limitations, Appendix A2 provides additional documentation on the construction of my dataset and the assumptions
behind it. The full coverage of the final sample is shown in Table A2.1, including details on which
country-year observations have been altered relative to their ’raw’ format. For verification, I crossreference my data with the Jordà et al. (2011) dataset, finding that where the two intersect, they
closely approximate each other. To link the Balance of Payments to the business cycle and financial
crises in the interwar period, it is supplemented with a host of other data, described in Appendix
A2-2. Summary statistics for the main interwar variables are shown in Table A2.2.
2.1.
The Balance of Payments
The Balance of Payments summarizes the transactions between residents and nonresidents over a
year. It is separated into the current and the capital account, whose balances (the difference between
credit and debit) are the inverse of each other, though small deviations due to accounting changes,
lagged payments or defaults are possible (IMF, 2009). Figure 1 shows capital account flows for the
US, Germany, the UK and Japan, and Figure A1.2 in the appendix reports the current account for the
same set of countries.6 A current account surplus is offset by a capital account deficit, indicating
increased claims on external financial assets. Consistent surpluses are consequentially equivalent to
an accumulation of net foreign assets. A consistent deficit, on the other hand, accumulates into net
foreign liabilities. The accumulated net international investment positions (NIIP) might then further
change due to either the revaluation or repayment of existing positions or by adding (subtracting) to
it via Balance of Payments flows (Lane and Milesi-Ferretti, 2004; Bleaney and Tian, 2013).
The NIIP can be separated into the gross international asset position (GIAP) and the gross
international liability position (GILP). Similarly, changes due to revaluations and repayments R N can
6As discussed in Appendix A2, the main limitation of the data is the uncertainty about the precise composition of gross
flows, with large heterogeneity across countries. The distinction between long-term and short-term flows, visible in the
figure, will consequently only be used descriptively or in robustness exercises, but not the baseline specifications.
7
be split into changes of gross assets (A) and gross liabilities (L), R N = R A − R L . Changes in the GILP
can consequently (analog to the NIIP) be separated into changes due to capital inflows (credit) and a
residual change due to revaluations and repayments of existing liabilities. Because cumulative flows
approximate changes in investment positions, less revaluations and repayments, the approximation’s
fit is largely determined by the magnitude of revaluations. When revaluations are cyclical, financial
flows provide close approximations, but when valuations steadily move in one direction, cumulative
flows grow gradually less precise (Atkeson et al., 2022). This relationship between the information
recorded in the BoP and international investment positions is expressed in Equation 1
Σnj=0 Creditt+ j = ∆n GILPt+n − Σnj=0 R L,t+ j ,
(1)
where Creditt refers to capital inflows, ∆GILP to the total change in the international liability position,
and R L to the residual change in the GILP due to revaluations and repayments.7 The left hand
side of Equation 1 is used as the main independent variable throughout the paper and, intuitively,
approximates changes in international investment positions over a given period.8 It is important
to emphasize that while cumulative current account balances approximate the NIIP, gross current
account flows do not accumulate into a stock of assets or liabilities and are not a theoretically
meaningful concept. First, because the items concerned are goods and services and not connected
to the acquisition of financial assets. Second, these items do not pile up into goods and service
positions, but the capital streams financing them potentially do. Third, already consumed goods are
not subject to revaluations. Additionally, variation in the current account must be driven by residents
of other countries deciding to purchase fewer, or more, goods from a particular country, a decision
unrelated to the gross financial flows attached to these transactions (Borio, 2016).
Sample Properties: In theory, the total amounts of worldwide credit and debit flows are always
equal, because the world is a closed financial system. Any sample not covering the entire world
or not representing a perfectly closed system might deviate from this parity. If a sample becomes
large enough to approach either condition, the difference between credit and debit will consequently
converge to zero. This also implies that the average net exposure to foreign capital will likewise
converge to zero. Figure 2 shows that this is the case for the group of countries covered by the
League of Nation’s BoP statistics. The left panel plots the credit and debit entries summed up over
all sample countries for the current account. The right panel does the same for the capital account.
In both cases the series for credit and debit almost perfectly mirror each other, as indicated by their
7 For the GIAP this is similarly defined as Σn Debit = ∆ GI AP
n
n
t
t+n − Σt R A,t , where Debitt refers to capital outflows,
t =0
∆GI AP to the total change in the international asset position, and R A to the change due to revaluations and repayments.
8 Because the historical BoP statistics reported by the League include all cross-border movements of financial assets
regardless of their origin, the above definitions differ slightly from modern definitions of foreign gross positions. For
example, the sale of either a domestic or foreign asset by a domestic to a foreign agent is considered a gross inflow in the
BoP (monetary inflow in exchange for asset ownership). In modern classifications, however, the sale of a domestic asset to
a foreigner is an increase in gross foreign liabilities, while the sale of a foreign asset is a decrease in gross foreign assets.
8
Figure 2: The balance of payments, sample properties
Billion US Dollars
0
0
1922
1926
1930
1934
1938
-10
-5
-20
-40
Balance
Inflows
Outflows
5
20
Billion US Dollars
10
Capital Account
40
Current Account
1922
1926
1930
1934
1938
Notes: This figure shows gross financial flows, summed over all sample countries, between 1922 and 1936. The left panel shows the
current- and the right panel the capital account. Inflows (credit) are plotted as positive values, outflows (debit) as negative values. Inflows
and outflows are shown in blue and red respectively. Their difference is shown in grey.
difference fluctuating around zero. This shows that the sample forms an almost closed system of
trade and capital flows, where gross flows are highly pro-cyclical, while aggregate net flows cannot
capture this dimension of capital movements. Additionally, the exercise serves as a validation of
the data, as persistent misreporting is unlikely to produce such a close match. Interestingly, capital
flows already peak in 1928, but do not shrink significantly until 1931. This relates back to this being
a period of capital flight and retrenchment, where new lending stops, but foreign capital is being
repatriated, resulting in large gross flow volumes in and out of countries.
3.
A short History of Capital Flows in the Interwar Period
This section takes a closer look at the development of capital movements in the interwar period. It
first highlights how key findings from the literature map into the Balance of Payments, followed by a
discussion of the major trends in interwar capital flows.
3.1.
From gold flows to net flows to gross flows
The interwar Gold Standard lacked the stability of the classical Gold Standard from its very beginning.
Many countries returned to gold on parities that no longer reflected their economic conditions,
disrupted by World War I, hyperinflation, and the unraveling of global trade (Eichengreen and Temin,
2000; Eichengreen, 2008; Irwin, 2012).9 The Gold Standard’s price-specie mechanism supposes that
countries with undervalued currencies attract gold, while countries with overvalued currencies will
9 Countries differed hugely in how and when it was implemented. Before WWI the Gold Standard was largely
homogeneous (Bordo and Kydland, 2005), but when countries returned to it (US in 1922, Germany in 1924, UK in 1925,
France implicitly in 1926 and explicitly in 1928 (Reinhart and Rogoff, 2009)), that changed. Some returned at overvalued
parities (US and UK), others at discounted rates (France) (Irwin, 2012). Some had large gold reserves (US), others almost
none (Germany) (Eichengreen and Mitchener, 2003). Most reinstated circulation of physical gold, but others (Germany)
never did (Deutsche Bundesbank, 1976). Some had accumulated gold in the 1920s (France), while others only had small
positive (US and Germany) or negative (UK) net inflows. The exit from gold was equally heterogeneous. Most countries
left gold during the Great Depression, but the gold block, led by France, accumulated enough gold to believe it could stay
on gold throughout the crisis, holding on until the mid-1930s (Bordo and Edelstein, 1999; Eichengreen and Irwin, 2010).
9
fail to do so (Bordo and Kydland, 2005; Wandschneider, 2008).10 The countries taking center stage
in this story are France, returning to gold at a vastly discounted rate, set in 1926 and formalized in
1928, and the United Kingdom returning to gold at the now overvalued pre-war parity in 1925. The
United Kingdom’s subsequent failure to attract gold resulted in it being the first major economy
to abandon gold and devalue its currency as early as 1931. France, instead, started to accumulate
gold from 1927 onward and remained on the Gold Standard until 1936. The other major economies
experienced gold inflows of smaller magnitude during the 1920s (Bernanke, 2009; Irwin, 2012).
These developments are shown in the left panel of Figure 3 using BoP data. It plots cumulative
net gold inflows since 1923, with France standing out as the largest importer of gold. This gold,
however, did not come by way of draining the other major economies of gold, but only hindered
them to accumulate as much gold themselves as they desired. The net gold inflow to France and,
in smaller magnitude, to Germany and the United States instead came from peripheral countries
(Eichengreen, 2008), as indicated by the gray line. In 1931, the United Kingdom, having failed to
attract gold in the 1920s, left gold and devalued its currency with immediate effect. Gold started to
flow into the country. In 1933, the US followed suit, devaluing the dollar and subsequently entering
a period of consistent gold inflows. This effectively appreciated the Franc, even though the Banque
de France fought to maintain the previous parity (Wandschneider, 2008; Irwin, 2012), with the result
that gold started to flow out of France starting in 1934 and before it left gold in 1936.
The return to gold was meant to be a signal for a return to the pre-war stability. Yet it also turned
out to be a facilitator of the external imbalances of the interwar years (Bordo and Kydland, 2005;
Wandschneider, 2008). These imbalances are manifested in the metaphor of a debt carousel in which
a group of creditor countries, centered on the United States, supplied a group of debtor countries,
centered on Germany, with credit (Kindleberger, 1978; Spoerer and Streb, 2013; End et al., 2019).
Foreign lending picked up steam during the Roaring Twenties when a common peg to gold ensured
predictable exchange rates and the free flow of capital. The result was an accumulation of imbalances
and an increasingly intertwined global ’web’ (De Broeck et al., 2018) of financial relations. The onset
of the Great Depression and the gradual abandonment of the Gold Standard broke this cycle, leading
to capital retrenchment and financial disintegration (Kindleberger, 1978; Bernanke, 2009).
The right panel of Figure 3 visualizes these developments by plotting the cumulative capital
account balance since 1923. It confirms the impression from Figure 2 that the sample approaches a
closed system where total net inflows equal net outflows. Three net creditors - the United States,
the United Kingdom and France - supply capital to the global financial system in general and to
Germany in particular. Interestingly, the Great Depression does not lead to a stagnation of net
10 The price-specie mechanism describes a self-balancing system in which, e.g., low price levels lead to trade surpluses
and an inflow of gold, which increases the money supply and raises prices, which ultimately balances the trade surplus.
During the interwar period, this mechanism became increasingly unreliable due to the sterilization of gold inflows (i.e.,
preventing the monetary base from expanding) and the transformation of the Gold Standard into an ”exchange” standard,
where countries could partially back their money in circulation with foreign assets in currencies convertible into gold.
10
Figure 3: Cumulative net gold- and net capital flows
-2
0
2
4
Cumulative Capital Imports in Billion $
1924
United Kingdom
Germany
-4
United States
France
All Other
-6
-6
-4
-2
0
2
4
Cumulative Gold Imports in Billio $
6
Capital
6
Gold
1927
1930
1933
1936
1924
1927
1930
1933
1936
Notes: This figure plots in the left panel the cumulative net gold inflows for the four major economies of the time, the US, UK, Germany
and France, as well as a fifth category including all other countries. Cumulative net inflows are plotted as positive values, cumulative net
outflows as negative values. The right panel plots the cumulative capital account balance for the same group of countries. Flows are
accumulated from 1923 onward, except for Germany, where the earliest available data is from 1924.
positions (as would be the expected result of a total breakdown in financial flows), but a reversal of
imbalances as creditors begin to repatriate their foreign assets. An exception is Germany which, after
settlements with its creditors and the rise of the National Socialists to power, stagnates on a high
level of net foreign credit (Ritschl, 2002a). Surprisingly, and against conventional wisdom, no sudden
stop or withdrawal of net foreign lending is discernible for the United States.11 In fact, the United
States continued to be a net capital exporter until 1933, reflecting the structure of its foreign assets,
which were mostly long-term and difficult to repatriate on short notice (Ritschl, 2002b).
It is reassuring that these narratives can be recreated using novel BoP data. But what explains the
focus on net flows inherent to both of them? Beyond the lack of suitable data, part of the reason may
lie in the very design of the system they aim to describe. The Gold Standard’s mandate for stability,
combined with the fact that the net availability of gold matters, resulted in a focus on ’balanced’
external flows (first discussed by Hume (1758) and well-documented in Eichengreen and Temin
(2000)). Following this tradition, Keynes (1941) linked countries’ net foreign deficits to contractionary
biases in the Great Depression,12 continuing a convention in which economies are characterized by
surpluses, deficits and imbalances, even though net flows are dwarfed by their gross counterparts.
But why should we care? As the next chapters will show, net flows often tell us little about gross
flows. For one thing, net flows might decrease while gross flows increase, but more often, they
simply do not grow by the same proportion (Borio and Disyatat, 2015; Borio, 2016). Focusing on net
flows consequently lets the volume of financial relationships go largely undetected. Only recently
has this come under scrutiny with Bernanke (2005) questioning the adequacy of the current account
11 For gross flows, the story is quite different, with a sudden stop occurring in 1928 (Accominotti and Eichengreen, 2016)
12 Determined to evade these imbalances in the future, he worked to enshrine rigid capital controls in the Bretton Woods
agreement in 1944, where again the overarching mandate was stability. Keynes advocated for the reduction of imbalances
by increasing the supply of international money. This effectively meant increasing the smaller of two gross positions to
decrease imbalances. The plan was abandoned in favor of a toned-down version with capital controls and ’special drawing
rights’ for net debtor countries (Crowther, 1949).
11
Figure 4: The relative size of gross capital flows: comparison to net flows, gold flows, and across countries
Debit Across Countries
1
Credit Across Countries
1
.5
Relative Sizes: Net and Gold Flows
United States
Germany
Netherlands
Other
1924
1927
1930
1933
1936
.8
United Kingdom
France
Canada
0
.2
.4
.6
.8
.6
.4
.2
0
0
.1
.2
.3
.4
Net / Capital
Gold / Current
1924
1927
1930
1933
1936
1924
1927
1930
1933
1936
Notes: The left panel of this figure shows the average ratio of absolute net capital flows to gross capital flows (blue) and, similarly, the
average ratio of gross gold flows to gross current account flows (gold) over time. The middle and right panel show countries’ shares in
total yearly credit (inflows) and debit (outflows) movements respectively. This is equivalent to the country-level time series for credit and
debit relative to the respective total flow volumes computed in Figure 2.
in describing the large global capital movements prior to the 2008 crisis, and Borio (2016) stating that
’current accounts have been asked to do too much and focusing on them excessively can lead policy astray’.
3.2.
Trends in the Balance of Payments
Studies on the Gold Standard or foreign imbalances are, by definition, concerned with net flows, as
net gold movements determine currency coverage and net capital flows shape imbalances. While
Figure 3 tracks both, it provides no information on the magnitude of capital flows in the interwar
period. To gauge these magnitudes, the left panel of Figure 4 plots average ratios of net to gross
capital- and gold to gross current account flows over time. Net flows fluctuate around 25% of gross
flows, reaching their trough in 1929, below 20%, when gross flows peak on the eve of the Great
Depression. The ratio of gold to gross current account flows stays below 5% through the 1920s until
it triples between 1930 and 1931, due to two factors: a sharp decline in gross current account flows as
global trade collapses, and the UK leaving the Gold Standard in 1931, devaluing its currency and
attracting large gold inflows. Figure A3.5 quantifies these observations using binned scatterplots
of the gold-to-gross capital and net-to-gross capital ratios against log gross capital flows. Both
relationships are distinctly negative, showing that when gross flows grow large, other types of flows
do not grow by the same proportion. Consequently, global uncertainty, capital retrenchment, and
financial fragility associated with international flows cannot be fully captured by net flows alone,
echoing the mixed success of net measures in linking capital flows to domestic economic and financial
conditions (Jordà et al., 2011; Gourinchas and Obstfeld, 2012; Mian et al., 2017; Kiley, 2021).
The middle and right panels of Figure 4 examine countries’ shares in total credit and debit flows,
revealing significant heterogeneity across countries and time. Stable shares indicate fluctuations
proportional to aggregate capital movements, while changes in shares suggest disproportionate
shifts. During the Great Depression, Germany and the United States experienced shrinking credit
12
shares, indicating larger reductions in inflows relative to the aggregate bust. Conversely, France’s
inflow share increased rapidly, suggesting it may have been perceived as a safe haven initially.13
US debit shares steadily increased prior to the Depression, accounting for about 40% of all capital
exports in 1928, but shrank to 20% in 1934. This shrinkage in capital exports, coupled with the
repatriation-driven increase in credit shares after 1933, explains the reversal in the US net position
shown in Figure 3. Similar patterns, though not always as clear, can be observed for other countries
in the sample, emphasizing the importance of studying gross flows to understand net flow dynamics.
Finally, Figure A3.6 in the appendix plots the sub-components of the current and capital account
individually. The boom-bust pattern around the Great Depression is now separated into different
account items rather than credit and debit. In the current account, most variation comes from trade
flows, followed distantly by services and secondary income flows (e.g., interest and dividends). Gold
accounts for the smallest fraction. As explained in Appendix A2, information on the composition of
gross capital movements should be interpreted cautiously. Nevertheless, Figure A3.6 reveals a sharp
change in capital flow composition around 1930: long-term flows plummet while short-term flows
increase, temporarily stabilizing total capital flows. Between the mid-1920s and mid-1930s, shortterm flows increase from about 30% to 50% of all flows, potentially reflecting increased economic
uncertainty and lending some credibility to their distinction in the data.
4.
Capital Flows and Business Cycle Dynamics
How do capital flows map into the business cycle and is it possible, despite the high aggregate
colinearity between credit and debit, to distinguish the implications of individual BoP components?
This section starts with local projections (Jordà, 2005) of GDP growth using BoP variables and then
continues by computing cumulative BoP positions, building on the intuitions developed in section 2,
to study the medium term relationship between capital flows and the business cycle.
4.1.
Output dynamics after Balance of Payments flows
To model the dynamic response of output following BoP flows I estimate local projections (Jordà,
2005) based on the following equation:
2
2
j =0
j =0
∆h yi,t+h = αi,h + ∑ βhC,j Crediti,t− j + ∑ βhD,j Debiti,t− j + γ X Xi,t + ui,t+h ,
(2)
where ∆h yi,t+h is log GDP growth14 for horizons h = 1, ..., 6 and Credit and Debit refer to the
corresponding items in the capital account. Since the Balance is a linear combination of gross flows
the response to it cannot be estimated in the same regression and is computed individually. All BoP
13 In fact, Baubeau et al. (2021) report that in 1930-1932 most foreign inflows were absorbed by french savings banks.
14 Log differences from all sources for GDP are combined to ensure maximum coverage. For transparency, the results for
sub-samples and the combined coverage are presented separately in Table A4.5.
13
Figure 5: Capital flows and business cycle dynamics
Year
4
6
.03
0
0
2
Year
4
6
-.06
2
Debit (ßD)
-.03
0
.03
Credit (ßC)
-.03
0
-.06
-.06
-.03
0
.03
Balance (ßB)
0
2
Year
4
6
Notes: This figure shows local projection results from Equation 2. The left panel plots the cumulative response of log GDP growth to the
capital account balance. The middle and right panel do the same for credit and debit respectively. The response to the capital account
balance (left) can be seen to be driven by the response to gross credit flows in the middle panel. Coefficients for credit and debit are
estimated jointly, but individually for the capital account balance. Standard errors are dually clustered on country and year. Shaded areas
represent 95% confidence intervals.
variables are normalized to unit standard deviation on the country level. Ultimately of interest are
the βh0 coefficients for balance, credit and debit over horizons h. All specifications control for two lags
of the independent variables and country fixed effects. The control vector Xi,t additionally includes
two lags of GDP growth in the baseline and additional controls in later robustness exercises.
The left panel of Figure 5 plots the response to the capital account balance. A one standard
deviation increase in the capital account balance (net inflows) is associated with a cumulative growth
slowdown of about 2.5 percentage points over the years t to t + 5. The response is statistically
significant at the 95% level across all horizons. Given that the capital account balance is simply the
inverse of the current account balance, these estimates directly map into findings where a deterioration
in the current account balance is linked to adverse outcomes (Jordà et al., 2011; Gourinchas and
Obstfeld, 2012; Kiley, 2021). The middle and right panel decompose the capital account balance
into its separate components by plotting the jointly estimated GDP responses to credit and debit
flows in year t respectively. The estimates show that the result in the left panel is driven entirely by
credit in the middle panel, while the response to debit is insignificant over all horizons. After a one
standard deviation increase of gross capital inflows in year t, cumulative GDP growth is reduced by
4 percentage points between t and t + 5.
Conceptually, this closely corresponds to the discussed BoP mechanics and the argument in Borio
(2016) and Borio and Disyatat (2015), where excess spending on goods and services, as captured in
the current account balance, needs to be financed with capital inflows from abroad. Over time these
inflows accumulate into foreign debt, but it is not the excess spending captured by net flows, but the
payment streams attached to them that ultimately matter for economic outcomes.
14
4.2.
BoP flows and business cycle dynamics in the medium term
Yearly flows have a measurable relation with future output dynamics, despite not accounting for the
potential accumulation of international investment positions over time. Yet, for capital inflows this
might be particularly important as they accumulate into foreign debt positions, which have been
linked to economic downturns empirically (Reinhart and Rogoff, 2009; Forbes and Warnock, 2012;
Caballero, 2016; Diebold and Richter, 2021), and theoretically (Schmitt-Grohé and Uribe, 2016; Mian
et al., 2020). Building on the intuition of section 2, I compute cumulative capital flows and estimate
the following predictive regression, similar to Mian et al. (2017) and Müller and Verner (2024):
∆2 yi,t+2 = αi + βhC
2
∑ Crediti,t− j
!
+ βhD
j =0
2
∑ Debiti,t− j
!
+ γ X Xi,t + ui,t+h ,
(3)
j =0
where ∆2 yi,t+2 is log GDP growth from year t to t + 2 in the baseline, and longer forecast horizons in
alternative specifications. BoP flows are summed over the three years from t to t − 2. All specifications
again control for country fixed effects and the vector Xi,t additionally includes two lags of GDP
growth in the baseline and additional controls in robustness checks. Due to every part of the BoP
being a linear combination of the other two, only two coefficients can be estimated jointly.
Columns (1), (4), and (7) in Table 1 report the coefficient for the cumulative capital account
balance for GDP forecast horizons t + 2, t + 3, and t + 4, respectively. It is significantly negative
across all specifications, confirming the dynamic response from the local projection exercise. Adding
cumulative gross foreign inflows in columns (2), (5), and (8) shifts predictive power away from net
inflows entirely. Both the coefficient and R2 increase twofold when gross capital inflows are included
in the model, while the coefficient for net inflows becomes close to zero and insignificant. The p-value
reported in the table consequently soundly rejects the equality of the two coefficients. Including both
types of gross flows in columns (3), (6), and (9) does not change the estimate for gross capital inflows,
which remains large and negatively significant. The coefficient for cumulative capital outflows is
zero. Again, the equality of coefficients can be rejected.
Along the time dimension the results in Table 1 are similar to the local projections in Figure 5.
Coefficients increase between the two and three year forecast horizon, but begin to phase out in
t + 4. This suggests that the response to capital inflows is concentrated in the first few years with
decreasing effects over time. As the sample is largest for the 2-year forecast horizon and the majority
of the response is concentrated in this period, this horizon is chosen as the baseline for the remaining
paper. All results, however, hold with alternative forecast lengths. Coefficients for cumulative credit
are notably larger (4 to 6 percentage points) than the previously estimated coefficients for yearly
flows (3 to 4 percentage points). This suggests that the effect of repeated gross foreign borrowing is
at least partially additive. When foreign credit accumulates, growth slowdowns become more severe.
15
Table 1: Capital flows and business cycle dynamics, 3-year cumulative capital flows
∆2 Yi,t+2
(1)
Σ2j=0 Balancei,t− j
-0.02∗∗∗
(0.01)
Σ2j=0 Crediti,t− j
(2)
(3)
-0.04∗∗∗
(0.01)
(4)
-0.03∗∗∗
0.01
(0.01)
Σ2j=0 Debiti,t− j
R2
Country fixed effects
Lagged Growth
p-value, β Credit = β Balance
p-value, β Credit = β Debit
Observations
∆3 Yi,t+3
(0.01)
-0.04∗∗∗
(0.01)
(5)
0.232
✓
✓
0.01
363
363
(6)
0.01
(0.01)
-0.06∗∗∗
(0.02)
0.00
(0.01)
0.123
✓
✓
∆4 Yi,t+4
(7)
(8)
-0.03∗∗∗
0.01
(0.01)
(0.01)
-0.05∗∗∗
(0.01)
-0.05∗∗∗
(0.02)
0.216
✓
✓
0.339
✓
✓
0.01
0.00
363
336
336
-0.05∗∗∗
(0.01)
-0.00
(0.01)
0.00
(0.01)
0.229
✓
✓
(9)
0.338
✓
✓
0.417
✓
✓
0.497
✓
✓
0.01
0.00
336
305
305
0.497
✓
✓
0.00
305
Notes: This table presents estimation results from Equation 3. The dependent variable is log GDP growth over horizons t to t + h. The
independent variables are cumulative capital account flows summed from t − 2 to t. All specifications control for country fixed effects.
Adjusting for longer time spans, lagged growth indicates two, three and four year distributed lags of GDP growth, respectively. The
reported p-values refer to a test for the equality of the computed coefficients. Standard errors in parentheses are dually clustered on
country and year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
4.3.
Robustness
How robust are these results and how do they compare to other explanations for the interwar
business cycle? Table A4.4 in the appendix first addresses the question of robustness over different
periods of capital flows. BoP variables are summed over five, instead of three years with results
remaining similar to the baseline. As this reduces sample size, but does not add much predictive
power to the model, the baseline specification of three-year BoP sums is employed throughout the
paper. Secondly, because GDP data is compiled from a variety of sources to ensure maximum
coverage, I show in Table A4.5 that it is not a sub-sample of GDP data driving the results. Coefficients
for the two largest contributors to GDP data and the total sample are estimated separately, with
coefficients being almost identical. Similarly, it might be possible that the aggregate results are driven
by large outliers. Figure A4.7 reports country-level coefficients for gross inflows, showing that the
negative relationship with GDP growth holds for the vast majority of countries.
Looking at the dynamic response of GDP in greater detail, Figure 6 re-estimates the local
projections, including additional control variables. For comparison, the baseline from Equation 2 is
plotted in blue. The specifications reported in orange and gold include the growth in domestic credit
and a Gold Standard indicator for years t to t − 2, respectively. This results in slightly dampened
coefficients but overall similar dynamics. The purple line includes net gold flows for the same three
years, with very little effect on the baseline coefficients. The same is true for the inclusion of a
financial crisis indicator, plotted in green. Together, these results confirm that the link between gross
foreign inflows and future growth dynamics is a consistent property of the data.
Similarly, Table A4.6 compares capital flows to other explanations of economic outcomes in the
interwar period by including additional explanatory variables first individually, and then jointly with
16
Figure 6: Capital flows and business cycle dynamics, robustness specifications
Year
4
6
.03
0
0
2
Year
4
6
-.06
2
Debit (ßD)
Baseline
Domestic Credit
Gold Standard
Gold Flows
Crises
-.03
0
.03
Credit (ßC)
-.03
0
-.06
-.06
-.03
0
.03
Balance (ßB)
0
2
Year
4
6
Notes: This figure shows local projection results from Equation 2 including additional control variables. The left panel plots the cumulative
response of log GDP growth to the capital account balance. The middle and right panel do the same for credit and debit respectively. For
the coefficients reported in orange the growth in domestic credit in years t to t − 2 is added to the baseline specification. The coefficients
in gold, purple and green include a Gold Standard indicator, gold flows and a financial crisis dummy for the same years respectively.
Standard errors are dually clustered on country and year. Shaded areas represent 95% confidence intervals.
capital flows. In line with Eichengreen and Mitchener (2003), I find that domestic credit growth is
significantly negatively related to future GDP growth. The same is true for a Gold Standard and a
financial crisis indicator variable. As it was arguably Gold Standard adherence during crises that
exacerbated downward pressure on the economy (Eichengreen, 1992), I also include their interaction.
While this produces a negative coefficient, it remains statistically insignificant. When including all
variables together with cumulative capital flows, gross foreign credit emerges as the most robust
predictor of economic downturns in the medium term. Interestingly, the interaction of foreign and
domestic credit also remains statistically significant, hinting at the potentially adverse effects of
foreign capital financing domestic credit (Caballero, 2016; Diebold and Richter, 2021).
Additional robustness checks, discussed in detail in Appendix A4, include sample splits (Table A4.7), year and country×period fixed effects (Table A4.8), scaling (instead of normalizing) capital
flow variables by GDP and bank balance sheet size (Table A4.9), and extending the analysis to
financial and non-financial equity returns as well as industrial production as alternative measures of
economic performance (Table A4.10). Across all specifications, gross foreign credit displays the most
significant relationship with adverse future outcomes.
5.
Capital Flows and Financial Fragility
Large international capital flows, especially inflows, often precede financial crises (Reinhart and
Rogoff, 2009; Caballero, 2016). These inflows, potentially unrelated to domestic conditions, can cause
maturity and currency mismatches and increase exposure to global uncertainty (Obstfeld, 2012; Rey,
2013). After crises, these flows tend to revert (sudden stops and capital flight) (Forbes and Warnock,
2012; Broner et al., 2013; Caballero and Simsek, 2020), as the cost of financial intermediation increases
(Jordà et al., 2013; Romer and Romer, 2017). Earlier work has linked current account balances,
17
particularly deficits, to crises (Caballero, 2016; Kiley, 2021). However, historically, crises seem just as
likely in surplus as in deficit countries (Obstfeld et al., 2010; Jordà et al., 2011). This chapter examines
the link between capital flows, financial crises, and post-crisis recession severity in the interwar
period, which, given its predominance, approaches a case study of the Great Depression.
5.1.
Capital flows and financial crises
I begin by establishing a descriptive link between capital flows and the frequency of financial crises,
primarily based on the crisis classification by Baron et al. (2021), supplemented by Reinhart and
Rogoff (2009) and Grossman (1994) for missing countries.15 Figure A5.8 shows the crisis probability
for different quintiles of cumulative capital flows from t − 2 to t. Crisis frequency in the highest
quintile of gross foreign inflows is about 15% (the highest of any quintile), but only 2% in the lowest
one. This pattern of increasing crisis frequencies from low to high quintiles is much less pronounced
for net inflows and gross outflows. To more formally exploit the connection between capital flows
and crisis occurrences I turn to a probit estimation, as it is widely used in the literature. Coefficients
are estimated based on Equation 4, where a financial crisis in country i in year t + 1 is denoted by
the indicator variable Fi,t+1 , conditional on capital flows from the Balance of Payments Xi,t
Pr [ Fi,t+1 = 1| Xi,t ] = Φ(βXi,t ).
(4)
Gross capital flows, as shown, are highly pro-cyclical, with the Great Depression dominating the
interwar crisis chronology. Due to this fact, the crisis prediction exercise relies on country-specific
gross capital imports and exports, as well as heterogeneity in the starting dates of crises (mainly
the Great Depression) across countries. The results are reported as mean marginal effects in Table 2.
Predictive accuracy is reported using the AUC statistic (Area Under the Curve), which is the integral
of the space under the ROC curve (Receiver Operating Characteristic) and the standard benchmark
for classification accuracy. An AUC of 0.5 indicates that the model’s choice of indicator variable is
random, while an AUC approaching 1 suggests the model effectively distinguishes between crisis
and non-crisis observations.
In line with the finding that crises are equally likely in surplus and deficit countries (Obstfeld
et al., 2010; Jordà et al., 2011), net capital flows in column (1) are not significantly related to future
crisis occurrences. However, together with the two included lags of GDP growth, the model has
some ability to sort the data into crisis and non-crisis bins, as indicated by an AUC of 0.62. Column
(2) adds gross capital inflows, which are significantly related to crisis occurrence. A one standard
deviation increase in gross inflows raises crisis probability by 0.6 percentage points. Given a sample
frequency of about 7%, this corresponds to crises being 9% more likely. The AUC increases to 0.7,
indicating improved precision in crisis identification. Column (3) includes both types of gross flows,
15 The final set of crises in the interwar period is described in Table A2.3.
18
Table 2: Capital flows predicting financial crises
Combined interwar crisis indicatori,t+1
Σ2j=0 Balancei,t− j
(1)
(2)
0.02
(0.02)
-0.02
(0.02)
Σ2j=0 Crediti,t− j
0.06∗∗∗
(0.02)
Σ2j=0 Debiti,t− j
AUC
s.e.
Lagged Growth
Country fixed effects
Observations
(3)
0.04∗∗∗
(0.02)
(4)
(5)
(6)
0.03
(0.03)
-0.03
(0.04)
0.05∗∗∗
(0.02)
0.09∗∗∗
(0.03)
0.01
(0.02)
(7)
(8)
0.07∗∗∗
(0.02)
0.07∗∗∗
(0.02)
0.02
(0.03)
0.62
0.05
✓
0.70
0.04
✓
0.69
0.04
✓
0.69
0.05
✓
382
382
382
382
0.70
0.05
✓
✓
233
0.76
0.04
✓
✓
233
0.75
0.04
✓
✓
233
0.75
0.04
✓
✓
233
Notes: The table shows estimation results of a probit model from Equation 4 for financial crises, reporting mean marginal effects. The
independent variables are cumulative capital flows from year t − 2 to t. AUC is the area under the ROC-Curve, below it is its standard
error. Lagged growth refers to a two year distributed lag of GDP growth. Columns (5) to (8) additionally include country fixed effects.
Standard errors in parentheses are clustered on country level and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
with gross outflows remaining insignificant. Finally, column (4) shows that the single-factor model of
gross inflows has the same predictive accuracy as the models in columns (1) to (3). In other words,
neither gross outflows nor net inflows add to the predictive power contained in gross capital inflows.
Since some sample countries, particularly developing economies or colonies, do not report any crises
for the sample period, columns (5) to (8) repeat the previous specifications with country fixed effects.
While the number of observations drops by a third, the results remain robust.
To check if the results are a feature of the crisis indicator, Table A5.11 repeats the exercise using
only the Reinhart and Rogoff (2009) crisis database. Although sample composition, number of
crises and individual starting dates differ, the results remain virtually identical. Together, these
results indicate that the information contained in gross foreign inflows best captures the capital flow
dynamics that have been observed in the run-up to financial crises.
5.2.
Increasing and decreasing recession severity: exposure to capital flows before crises
Gross foreign inflows are predictive of financial crises, but does the degree to which countries were
exposed to international capital movements also matter for the severity of the post-crisis recession?
The idea that large foreign borrowing before crises might turn out to be harmful afterward was
already explored by Keynes (1941), who linked capital retrenchment to contractionary biases in
net debtor countries. Further studies, however, explaining the severity of the Great Depression
conditional on international capital flows are scarce. In the following, I provide empirical evidence
that large exposure to gross foreign inflows before crises was followed by more severe recessions,
while the accumulation of foreign assets provided some protection against this mechanism.
Exposure to capital inflows: Jordà et al. (2013) show that countries have deeper post-crisis
recessions when the preceding boom in domestic credit was large. Borio et al. (2014) confirm this in
19
Figure 7: Recession severity after exposure to gross capital inflows
Low exposure crises
.2
High exposure crises
.2
.2
All Crises
Year
4
6
.1
0
2
Year
4
6
-.2
2
-.1
0
.1
0
-.1
0
-.2
-.2
-.1
0
.1
Average Cumulative Growth
Local Projection Coefficient
0
2
Year
4
6
Notes: This figure shows cumulative GDP growth over a six-year window following different types of financial crises. The left panel
displays the average across all crises, which are split into crises with large prior exposure to foreign inflows in the middle panel and all
other crises in the right panel. High exposure crises are defined by the interaction of the GED measure from Equation 5 with a financial
crisis indicator. The average cumulative GDP growth after the different crises is plotted in red. Plotted in blue are estimates based on a
local projection of GDP growth on an indicator variable for the different types of crises in year t, including country fixed effects and two
lags of GDP growth and crises. Shaded areas represent 90% confidence intervals.
a case study of the Great Depression and argue that these countries were linked via large capital
flows. Figure A6.9 in the appendix approaches this idea descriptively for foreign credit, plotting
gross foreign inflows from 1927 to 1930 against log GDP growth from 1930 to 1933, which produces
a visibly negative relationship. This, however, does not account for country-specific starting points
of the Great Depression. The right panel addresses this concern by splitting the sample along the
median recession severity in the first three years after a crisis, similar to Borio et al. (2014). Plotting
average gross capital flows for both groups in a six-year window around crises reveals that countries
where the post-crisis recession was deeper than the median decline consistently had higher gross
inflows before crises but experienced a sharper contraction in foreign inflows afterward.
This finding corresponds to literature suggesting that sudden stops and capital flight occur
after periods of elevated capital flows and that the sudden unavailability of funds during crises is
potentially harmful to the economy (Reinhart and Rogoff, 2009; Forbes and Warnock, 2012; Broner
et al., 2013). To take a more systematic look at the economic development of countries that were
exposed to large levels of gross capital inflows before crises, Equation 5 defines exposure as large
when gross inflows were above the yearly median in two consecutive years (t − 1 and t − 2)
GEDi,t =

1,
^
^
if Crediti,t−1 > Credit
i,t−1 ∧ Crediti,t−2 > Crediti,t−2
0,
Otherwise.
(5)
Interacting the gross exposure dummy (GED) with a crisis indicator for time t identifies crises
with high previous exposure to gross inflows. This classification applies to about 25% of crises in the
sample. Figure 7 shows that GDP growth after such an inflow-crisis is much lower than for other
crises over the six-year window following the beginning of the crisis. Importantly, this relationship
is visible in the purely descriptive exercise of computing the average cumulative GDP growth after
20
different types of crisis, plotted in red. Local projections, plotted in blue, allow me to control for
country fixed effects and two lags of GDP growth and financial crises. This confirms the previous
results, with the cumulative growth coefficients for GED-crises being significantly lower than for
Non-GED-crises at the 90% level at horizons 3 and 4.
To test the relationship between crises, foreign inflows and recession severity more comprehensively, Equation 6 runs predictive regressions of GDP growth on a crisis indicator interacted with the
previously constructed GED measure
∆2 yi,t+2 = αi + βCr Crisisi,t + βGED GEDi,t + βCr×GED Crisisi,t × GEDi,t + γ X Xi,t + ui,t+2 ,
(6)
where the dependent variable ∆2 yi,t+2 is log GDP growth from t to t + 2. Column (1) in Table 3
reports coefficients for crises and the GED-variable individually. The coefficient for crises is negative,
while it is zero for the GED-measure. When the two are interacted in column (2), the picture is
strikingly different. The interaction coefficient is larger than the one for crises and significantly
negative. To confirm that the interaction captures the state-dependent implications of financial crises
and does not proxy for the previously documented negative link between gross inflows and growth,
column (3) adds the baseline BoP variables. The interaction coefficient remains unchanged, while
the gross inflow coefficient is identical to the baseline. This suggests that relying on foreign credit
before crises exacerbates the negative association that persists between gross inflows and economic
outcomes across all specifications. Column (4) adds a Gold Standard indicator to control for the
potential effects of restrictive monetary policy during crises, which does not change either of the
coefficients. Some countries - particularly colonies and developing economies - do not report any
crises for the sample period. To ensure that the interaction does not capture countries without crises
not being exposed to foreign inflows, columns (5) to (8) repeat the previous specifications but restrict
the sample to countries that report at least one crisis. The results remain unaffected.
Robustness: Is this result driven by the dummy classification or the choice of crisis chronology? I
turn the question upside down and define the dummy not as the relatively small sample with high
exposure in two consecutive years, but instead simply take the first lag of my baseline variable (such
that it does not overlap with a potential crisis in t) and define exposure as high if it is in the top 80%
of the entire sample. When interacted with a crisis indicator this effectively excludes countries with
low exposure (the bottom 20%), while capturing the vast majority of crises. The result is reported
in Table A6.12 and shows that focusing on the exclusion of low exposure countries produces very
similar results. In Table A6.13 I re-estimate Table 3 using the Reinhart and Rogoff (2009) crisis dating.
The results remain unchanged. Together, these result indicate that exposure to gross inflows prior to
crises adds to the negative link between gross inflows and economic outcomes.
Accumulating foreign capital: The accumulation of foreign debt in one country implies the
accumulation of foreign assets in another. Similarly, capital flight during crises implies the repatriation
21
Table 3: GDP growth, crises and exposure to gross capital inflows
∆2 Yi,t+2
(1)
(2)
Crisisi,t
-0.05∗∗
-0.02∗
(0.02)
(0.01)
GEDi,t
-0.01
(0.01)
(4)
(5)
(6)
(7)
(8)
-0.01
(0.02)
-0.00
(0.02)
-0.05∗∗
-0.02∗
(0.02)
(0.01)
-0.01
(0.02)
-0.01
(0.02)
-0.00
(0.01)
0.03∗∗
(0.01)
0.02∗
(0.01)
-0.02
(0.02)
-0.01
(0.02)
0.02
(0.02)
0.02
(0.02)
-0.08∗∗∗
(0.03)
-0.07∗∗∗
(0.02)
-0.07∗∗∗
(0.02)
-0.08∗∗
(0.03)
-0.07∗∗∗
(0.03)
-0.07∗∗∗
(0.03)
Σ2j=0 Crediti,t− j
-0.04∗∗∗
(0.01)
-0.04∗∗∗
(0.01)
-0.04∗∗∗
(0.01)
-0.04∗∗∗
(0.01)
Σ2j=0 Balancei,t− j
0.00
(0.01)
0.00
(0.01)
0.00
(0.01)
0.00
(0.01)
Crisisi,t × GEDi,t
(3)
-0.03∗
(0.02)
Gold Standardi,t
R2
Country fixed effects
Lagged Growth
Crisis in Sample
Observations
0.093
✓
✓
0.111
✓
✓
0.271
✓
✓
0.290
✓
✓
342
342
342
342
-0.02
(0.02)
0.126
✓
✓
✓
241
0.146
✓
✓
✓
241
0.325
✓
✓
✓
241
0.330
✓
✓
✓
241
Notes: This table presents estimation results from Equation 6. The dependent variable is log GDP growth over the period t to t + 2. The
independent variables are a financial crisis indicator, the GED-variable capturing exposure to large capital inflows, the baseline BoP
variables accumulated over t to t − 2 and the Gold Standard. All specifications additionally control for country fixed effects and a two
year distributed lag of GDP growth. Columns (5) to (8) restrict the sample to countries that report at least one crisis episode. Standard
errors in parentheses are dually clustered on country and year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
of foreign assets by other countries, which consequently experience capital inflows (potentially even
outweighing the flight of foreign capital). While gross foreign lending decreases, these countries may
cushion the effect on their economies via the repatriation of their own foreign assets (Caballero and
Simsek, 2020). Contrary to recessions being more severe due to foreign credit exposure, this channel
proposes the moderation of recessions via the liquidity insurance provided by foreign assets.
In Table A6.14, I define a dummy for accumulated gross foreign assets (GFA) analogously to
the GED measure, taking the value 1 if gross outflows were above the yearly median in t − 1 and
t − 2. Its interaction with crises is positive, providing evidence for the hypothesis that foreign assets
may dampen crisis effects. Assuming the pre-crisis boom splits countries into capital exporters and
importers, GFA and GED might capture the same information, showing on the one hand that gross
importers fare worse and on the other that exporters fare better. Including capital flow variables
and the GED-crisis interaction, however, reveals these channels to be independent of each other.
This corresponds to the model in Caballero and Simsek (2020), where a country may be adversely
affected by inflow exposure, but simultaneously benefit from owning foreign assets. To verify that
it is not ultimately the net availability of funds that matters during a crisis, I interact a dummy for
positive net inflows (NID) in year t with the crisis indicator. The coefficient is positively significant,
confirming that the net availability of funds may help moderate crises, but does not affect the other
coefficients. Finally, I interact crises with a Gold Standard indicator, showing neither result proxies
for Gold Standard adherence, and replicate the results in a sample of countries that report at least
one crisis, again with unchanged results.
22
The positive effect of foreign assets is generally of lower magnitude and significance than the
negative effect of foreign liabilities. Assuming perfect substitution between foreign assets and
liabilities and considering their identical volumes across the sample, their coefficients should offset
each other. A potential explanation for the discrepancy lies in why foreign funds return home
during crises. For repatriated assets to fully counter withdrawn foreign funds, all repatriated assets
would need to be re-employed domestically. This is unlikely, as theory suggests that the need
for precautionary savings increases during crises, depressing aggregate demand (Guerrieri and
Lorenzoni, 2017). Empirically, Degorce and Monnet (2020) confirm that reluctance to invest or
consume led to increased precautionary savings during the Great Depression. In the context of
capital flows, foreign assets repatriated due to increased caution might only partially be channeled
back into the domestic economy, with the rest held in savings accounts. This may cushion the shock
of foreign capital withdrawal but cannot fully substitute for it.16
5.3.
Crises, capital flight and sudden stops
Reliance on capital inflows amplifies the ramifications of financial crisis, and, importantly, adds to
the negative relationship between gross inflows and future output growth. A closer examination
of post-crisis capital movements helps to explain these findings. Once a crisis occurs, uncertainty
in financial intermediation increases, leading to stops in lending, capital flight and contractions in
flow volumes,17 and, in the context of the Great Depression, the overseas-disinvestment of the US
exacerbated the crisis by causing liquidity problems abroad (Kindleberger, 1978; Eichengreen, 2008).
Intuitively, the reduction in foreign funding disproportionately affects countries previously reliant on
foreign capital. To explore capital flow dynamics after crises, I estimate the following specification
n
2
h =1
j =0
X
∑ Capitali,tT +h = αi,h + ∑ βT,h
Cr,j Crisisi,t− j + γ Xi,t + ui,t+h ,
(7)
T
where ∑nh=1 Capitali,t
+h are cumulative capital flows of type T ∈ { Balance, Credit, Debit } from year
t to t + h. Flows are split into long and short-term flows to emphasize the difference in response
time to the crisis. The β T,h
Cr,0 coefficient measures the response of the respective flow type to a crisis in
year t over the various horizons h. All specifications include country fixed effects and two lags of the
crises variable. The vector Xi,t contains two lags of GDP growth and capital flows.
The left panel of Figure 8 plots the response of net capital flows. The short-term capital balance
immediately drops in the first year after crises, indicating a net decrease in short-term capital, and
remains significantly negative over all horizons. The change in long-term flows is not significantly
16 This argument is consistent with the ’paradox of thrift’ (Keynes, 1936), where higher savings decrease output growth
via depressed aggregate demand, leading to the paradox of increased savings today lowering total savings in the long run.
17 See Jordà et al. (2013); Krishnamurthy and Muir (2017); Romer and Romer (2017) for disintermediation and (Calvo,
1998; Forbes and Warnock, 2012; Broner et al., 2013) for decreasing capital flows after crises.
23
Figure 8: Cumulative capital flows after crises
B
C
D
0
-2
-1
0
-1
-2
-2
-1
0
1
Debit Response (ß )
1
Credit Response (ß )
1
Balance Response (ß )
2
Year
4
6
0
2
Year
4
6
-3
0
-3
-3
Long
Short
0
2
Year
4
6
Notes: This figure shows local projections of cumulative capital flows following financial crises, based on Equation 7. The left panel plots
the response of the capital account balance, split into long- and short term capital flows. The middle and right panel do the same for
credit and debit respectively. Standard errors are dually clustered on country and year. Shaded areas represent 90% confidence intervals.
different from zero. This response is a combination of a decrease in gross inflows (sudden stop) and
an increase in gross outflows (flight). Gross inflows consistently trend downward, with long-term
flows naturally taking longer to react than short-term flows. The response for outflows is less
clear-cut but trends in the opposite direction after crises, with especially short-term flows being
elevated in t + 1 indicating capital flight from crisis countries.
These findings relate to Accominotti and Eichengreen (2016), who term the contraction in foreign
lending during the Great Depression the ”mother of all sudden stops”. Interestingly, they time the
reversal from 1928 to 1929, while my results rely on a heterogeneous crisis indicator across countries.
Crucial to their timing is a change in the relationship between the world’s largest net creditor and
debtor. Under the transfer protection clause of the Dawes Plan, foreign investors had privileged access
to German debt payments during crisis events. In early 1929, the Young Plan effectively reversed
this seniority by establishing that reparations had to be paid under any circumstances, significantly
increasing the risk for foreign investors (especially private US creditors), and triggering a sharp
reduction in US capital exports (Ritschl, 2012; Ritschl and Ho, 2023). My results expand on these
findings, showing that while the initial reversal may have occurred between 1928 and 1929, individual
countries’ crisis starting points remain crucial for understanding capital flow dynamics.
5.4.
Capital flow episodes
The cyclicality of capital flows around the Great Depression allows me to consider their relationship
with financial fragility separately during booms and busts, while also addressing a data caveat.
Following Forbes and Warnock (2012), I categorize the data into four discrete types: surges, stops,
flight, and retrenchment. Surges (stops) are observations for which the change in gross inflows lies one
standard deviation above (below) its ’historic’ mean, defined as the rolling average of the previous
24
Table 4: Capital inflow episodes and financial fragility
Episode
Reg. Sample
Surge
Surge
Stop
Stop
Level
All
High
Low
High
Low
N
363
21
42
31
31
Year
1931.2
1930.5
1934.7
1930.6
1932.6
GDP
Crises
∆2i,t+2
∆3i,t+3
∆4i,t+4
Ci,t
Ci,t+1
Ci,t+2
0.041
-0.004
0.093
-0.032
0.052
0.061
-0.009
0.158
-0.027
0.083
0.078
0.018
0.207
-0.000
0.125
0.090
0.192
0.063
0.143
0.059
0.086
0.080
0.042
0.176
0.088
0.089
0.043
0.023
0.156
0.063
Notes: This table shows GDP growth and crisis likelihood following capital inflow episodes, combined with the levels at which they occur.
’Surges’ (’stops) are observations where the current change in gross inflows lies one standard deviation above (below) its ’historic’ mean,
defined as the rolling average of the previous three years. Levels are defined relative to the country-level median. For comparison, the first
row reports the same statistics for the regression sample.
three years. Flight and ’retrenchment’ apply the same logic to outflows.18 What these episodes
capture changes fundamentally around the Great Depression. During the 1920s, surges in inflows
reflect increased foreign lending or investments, which seems unlikely in the early 1930s when they
presumably capture increased capital repatriation. This complicates assessing whether the effect of
inflows is symmetrical (i.e., if increasing inflows relate to lower growth, do decreasing inflows relate
to higher growth?), as the answer is likely state-dependent. In Table 4, I consequently report statistics
for GDP growth and crisis likelihood following changes in inflows in combination with the level at
which they occur, distinguishing high and low levels based on the country-level median.
Row (2) shows the combination of a surge in capital flows when levels are high, occurring on
average in mid-1930. It is followed by GDP growth far below and crisis likelihood at time t far above
the sample mean, shown for comparison in row (1). In contrast, row (3) shows that increasing inflows
when levels are low are followed by higher than average growth and decreased crisis likelihood.
Given previous results, these inflows likely represent returning capital exports of the 1920s (or
reopening to international markets during recovery). Row (4) demonstrates that a stop in inflows
when levels are high relates to the worst economic performance in the following three years, with
elevated crisis likelihood in t, t + 1, and t + 2, corresponding to findings where sudden stops in
inflows trigger crises (Reinhart and Rogoff, 2009; Forbes and Warnock, 2012). Further reductions
when levels are low (row 5) are closest to the sample average. These findings indicate that the baseline
result may underestimate the effect of foreign borrowing, as it also includes the positive implications
of having access to international markets and the repatriation of foreign assets during crises. I will
revisit this point in a later IV-specification, shutting down this channel by exogenizing foreign capital
supply. For completeness, the results for outflow episodes are reported in Table A6.15. While not
being as striking (nor as diverse) as the results for inflows, they generally align with intuition.19
18 Time windows are shortened compared to Forbes and Warnock (2012), yet this still induces a bias towards a later
dating of episodes, as coverage generally starts in 1923 or 1924, and at least 3 periods are required for the historic mean.
19 E.g., flight or retrenchment at high levels is associated with increased crisis likelihood. The ambiguity of the flight
indicator (which may reflect foreign asset purchases) becomes evident at low levels, where growth exceeds the mean.
25
6.
Channels: Connecting capital flows to real outcomes
The previous chapters have shown that gross foreign inflows are robustly associated with adverse
economic outcomes. But through which mechanisms does this happen? To address this question,
this section first investigates how capital inflows surged following the adoption of the Gold Standard
and how the Gold Standard simultaneously constrained the tools available to manage them. Subsequently, it examines the connection between foreign inflows and financial fragility through maturity
mismatches on commercial bank balance sheets. It then considers that borrowing from abroad today
comes with interest payments to foreigners tomorrow, showing that these interest payments are
negatively related to future growth prospects.
6.1.
The Gold Standard
Global financial integration: The Gold Standard was a commitment to the free flow of capital,
reduced risk in international lending relations and fostered financial integration via the common
peg to gold (James, 1992; Bordo and Kydland, 2005; Wandschneider, 2008; Eichengreen, 2008). While
this facilitated access to international capital markets, it also potentially contributed to increased
exposure to the dynamics described above. To judge the contribution of Gold Standard adoption to
the boom in international capital flows, Equation 8 estimates local projections of capital flows on an
indicator for Gold Standard adoption
n
2
h =0
j =1
T,h
T
X
∑ Capitali,tT +h = αi,h + βT,h
G GS Adoptioni,t + ∑ β j Capitali,t− j + γ Xi,t + ui,t+h .
(8)
T
Here, ∑nh=0 Capitali,t
+h are cumulative capital flows of type T ∈ Balance, Credit, Debit from year
t to t + h. GS Adoption refers to an indicator that identifies the first year a country is on the Gold
Standard. Ultimately of interest is the β T,h
G coefficient for all capital flow types across horizons
h = 0, ..., 6. Estimations for gross capital flows include two lags of credit and debit, while estimations
for net flows include two lags of the balance. The vector Xi,t contains two lags of GDP growth.
The results in blue in Figure 9 show that net capital inflows increase upon Gold Standard adoption
(left), and that this effect is exclusively driven by increasing gross inflows (middle). The increase is
statistically significant after h = 2, and continues until h = 4. Interestingly, the response for gross
outflows (right) is insignificant across all horizons, as capital exports likely depend on other countries
adopting the Gold Standard and becoming potential destinations for capital exports. To account for
the possibility that inflows simply grow proportionally to GDP after Gold Standard adoption, I shift
my control vector Xi,t (containing two lags of GDP growth) forward with the forecast horizon h and
plot the results in red. Strikingly, the results remain unchanged, indicating that the increase in capital
inflows due to Gold Standard adoption goes beyond what can be explained by GDP growth.
26
Figure 9: Gold Standard adoption and cumulative capital flows
B
C
D
0
2
Year
4
6
2
-2
-2
-2
Baseline
Shifted GDP growth
0
2
0
0
2
4
Debit Response (ß )
4
Credit Response (ß )
4
Balance Response (ß )
0
2
Year
4
6
0
2
Year
4
6
Notes: This figure shows local projections of cumulative capital flows to an indicator for Gold Standard adoption, based on Equation 8.
The left panel plots the response of the capital account balance, the middle and right panel do the same for credit and debit respectively.
The response plotted in blue corresponds to the baseline specification. The response plotted in red shifts lagged GDP growth together
with the forecast horizon. Responses in the balance are estimated individually and then decomposed into jointly estimated responses of
credit and debit. Standard errors are dually clustered on country and year. Shaded areas represent 90% confidence intervals.
Domestic Policy: The international dimension of the Gold Standard enabled financial integration
and increased exposure to foreign capital inflows. However, it was the domestic constraints imposed
by the Gold Standard that limited policy responses to these capital movements. Under the open
economy trilemma, the Gold Standard restricted domestic policymakers’ ability to counter the effects
of inflows through monetary policy, capital flow management, or exchange rate adjustments. This
created a situation where the rules of the Gold Standard exposed countries to global credit but left
them unable to respond adequately. Arguably, this makes the trilemma a key factor in understanding
the interwar crisis, as it reflects the ’inability to pursue consistent policies in a rapidly changing economic
environment’ (Obstfeld et al., 2004). To estimate the protection provided by being off the Gold
Standard, column (1) of Table 5 interacts a dummy for not being on gold between t − 2 and t with the
baseline Credit variable. The interaction coefficient is positive, only slightly smaller than the negative
Credit coefficient, and statistically significant. In columns (2) and (3), I control for net inflows and
gross outflows, respectively, which does not alter the results. This suggests that the ability to react to
foreign inflows matters for the magnitude of their impact.
Davis et al. (2016) emphasize that an open capital account is important for current account deficits
(a net measure) to have adverse domestic effects, and Quinn (2003) finds evidence that countries
with more open capital accounts had deeper recessions during the Great Depression. Following this
reasoning, I interact gross inflows with a dummy for the capital account being less than completely
open (below 100%) between t − 2 and t in column (4).20 Again, the interaction coefficient is positive
and statistically significant. In column (6), I repeat the specification for further reduced capital account
openness (below 66%), resulting in the interaction coefficient slightly increasing in magnitude. This
implies that capital account management may help moderate the negative implications of foreign
20 I use the capital account openness measure from Quinn (2003), which is highly correlated with the Gold Standard.
Among Gold Standard observations, the average openness score is 93, compared to only 49 outside of it.
27
Table 5: Capital flows, the Gold Standard and capital account openness
∆2 Yi,t+2
Σ2j=0 Crediti,t− j
Σ2j=0 Crediti,t− j × No Goldi,t→t−2
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
-0.04∗∗∗
-0.04∗∗∗
-0.05∗∗∗
-0.06∗∗∗
-0.07∗∗∗
-0.06∗∗∗
-0.06∗∗∗
-0.06∗∗∗
(0.01)
(0.02)
(0.01)
(0.02)
(0.02)
(0.02)
(0.02)
(0.02)
-0.06∗∗∗
(0.02)
0.03∗∗
(0.02)
0.03∗∗
(0.02)
0.03∗∗
(0.02)
0.04∗∗
(0.02)
0.04∗∗
(0.02)
0.04∗∗
(0.02)
0.05∗∗
(0.02)
0.05∗∗
(0.02)
0.05∗∗
(0.02)
0.380
✓
✓
0.381
✓
✓
✓
0.380
✓
✓
223
223
Σ2j=0 Crediti,t− j × Closed (< 100)i,t→t−2
Σ2j=0 Crediti,t− j × Closed (< 67)i,t→t−2
R2
Country fixed effects
Lagged Growth
Net Capital Inflows
Gross Outflows
Mean Σ2j=0 Crediti,t− j | ind. = 0
Mean Σ2j=0 Crediti,t− j | ind. = 1
Observations
0.274
✓
✓
.36
-.42
342
0.274
✓
✓
✓
342
0.277
✓
✓
0.407
✓
✓
✓
342
.37
-.2
223
0.409
✓
✓
✓
223
0.407
✓
✓
✓
223
.3
-.31
223
✓
Notes: This table presents results from regressing log GDP growth from t to t + 2 on interactions of gross capital inflows with measures of
Gold Standard adherence and capital account openness. Columns (1) to (3) interact gross inflows with a dummy for not being on gold
between t and t − 2, columns (4) to (6) with a dummy for the capital account being less than 100 % open (based on Quinn (2003)), and
columns (7) to (9) repeat the specification for further reduced capital account openness (below 66%). The mean of 3-year gross inflows for
the respective indicator variable being 0 or 1 is shown when indicated (for normalized variables, larger negative values represent lower
inflows). All specifications include the gold and capital account variables individually, country fixed effects, and two lags of GDP growth.
Standard errors are dually clustered on country and year and *,**,*** indicate significance at the 0.1, 0.05, and 0.01 levels, respectively.
inflows. The similarity in results between the Gold Standard and the capital account measure also
relates to Mitchener and Wandschneider (2015), where the authors find that countries introduced
capital controls after leaving the Gold Standard, while the option for independent monetary policy
was underutilized. For robustness, Table A7.16 reports the full set of interactions for all included BoP
variables with the various openness measures. The results remain unchanged.
Reserve accumulation: Another important dimension of both the Gold Standard and capital flows
are official reserves, which during the interwar period effectively meant physical gold or foreign
assets. When adopting or leaving the Gold Standard, capital flow measurements might have been
distorted by central banks reallocating official reserves between gold and foreign assets.21 On the
other hand, capital flows are an important determinant in the accumulation (depletion) of official
reserves (Pineau et al., 2006), potentially affecting the economy through them. Consequently, capital
flows and changes in official reserves are often considered distinct items in modern studies (Forbes
and Warnock, 2012; Broner et al., 2013). As the BoP does not report them individually, I collect
additional data on official reserves and, to assess the extent to which capital flows might proxy for
changes in official reserves, estimate their relationship with capital flows in Table A7.17.22 The results
21 Though coming under the same headline of official reserves, gold and foreign assets relate to different flows within the
BoP. Gold is recorded in the current account, while foreign assets are recorded in the capital account. E.g., selling physical
gold in exchange for foreign currency is recorded as a current account export (monetary inflow, outflow of goods). The
purchase of foreign assets on financial markets, however, is recorded as an outflow in the capital account.
22 The data for gold and foreign asset reserves is compiled from League of Nations (1936-1939b) and League of Nations
28
show that foreign credit (net and gross) is associated with decreasing foreign asset reserves, while
capital exports are linked to increasing foreign reserves. For gold reserves, the relationship is weaker
(with reversed signs), as capital flows are linked to gold movements only indirectly (mechanically
through the BoP). Reassuringly, even this weaker relationship disappears when physical gold flows
from the current account are included in the model.
To judge the economic dimension of this relationship, I include changes in official reserves
and capital flows separately in predictive regressions of GDP, mirroring my baseline specification.
Table A7.18 shows that increases in foreign asset reserves are followed by higher GDP growth, with
coefficients growing in magnitude and significance over longer horizons. Changes in gold reserves,
however, are not related to economic outcomes.23 Table A7.19 consequently focuses on foreign asset
reserves, interacting them with an indicator for financial crises. This shows that higher levels of
foreign asset reserves helped moderate post-crisis recession severity. The interaction coefficient
increases in magnitude and statistical significance when the United States, an outlier in terms of
both foreign asset holdings and recession severity, is excluded from the sample. Together, these
results help to further rationalize the previous finding of capital inflows aggravating and exports
moderating recession severity, suggesting a channel where capital inflows are linked to the economy
through their relation with official reserves.
6.2.
Bank balance sheet expansions and maturity mismatches
Bank balance sheets expand during period of growth when asset valuations rise and capital flows
into the financial system, but contract during downturns when asset values decline and capital is
withdrawn. During downturns, financial institutions with insufficient short-term assets are often
forced to resort to rapidly liquidating illiquid (long-term) assets to service the withdrawal of liquid
(short-term) liabilities, exacerbating asset price declines and hastening capital withdrawals (Adrian
and Shin, 2010). Beyond short-term ”fire sales”, continued contractionary pressure on bank balance
sheets leads to reduced credit creation, adversely affecting future real economic outcomes (ChodorowReich, 2014). When balance sheet expansions have been foreign-financed, vulnerability to sudden
reductions in capital availability increases, as foreign inflows typically have shorter maturities than
the domestic investments they fund (e.g. household or corporate credit). Importantly, these maturity
mismatches extend beyond the marginal (net) inflow of foreign capital. Instead, balance sheets may
expand due to foreign financing in both surplus and deficit countries, as countries can simultaneously
(1921-1939c), which record gold-to-notes-in-circulation, gold-and-foreign-assets-to-notes-in-circulation, as well as gold
reserves and notes in circulation individually. Ideally, capital flows from the BoP could be ’cleaned’ of official reserves;
however, as changes in reserves represent a net change between two periods, I cannot directly link these to gross flows.
23 Explanations include (i) that physical gold lost in relative importance in the interwar Gold Exchange Standard, (ii) gold
reserves became largely irrelevant when a country left gold, while foreign asset reserves could be used under any regime,
(iii) increases in the gold ratio when below the coverage threshold should not affect policy or the monetary base, and (iv)
expansionary effects were potentially offset by countries with large gold reserves (the Gold Block) staying on gold for
longer and remaining ’fettered’ (Bordo and Edelstein, 1999; Eichengreen and Irwin, 2010).
29
be international investors and recipients of foreign investment. In the event of a sudden capital
withdrawal, even financial systems in countries with a surplus of foreign assets and sufficient liquid
assets at the national level may be vulnerable, because the agent holding foreign assets may not be
the same as the one facing the withdrawal of foreign capital (Obstfeld, 2012).24
To test this intuition, I link capital inflows from the BoP to newly collected bank balance sheet
data from the League of Nations (1931-1940). These balance sheets contain a breakdown into
deposits, debt instruments (cheques, drafts, acceptances, re-discounts, interbank credit), reserves,
and ’other instruments’ on the liability side, and loans, bonds, securities (excluding bonds), and
’other instruments’ on the asset side. Relying on the method used in Fang et al. (2022) and Eren et
al. (2023), Equation 9 decomposes the relationship between total bank balance sheet size BBSi,t and
capital flows into the individual linkages between capital flows and particular instruments, such that
the instrument specific coefficients add up to the coefficient for the total balance sheet
j
j
BBSi,t − BBSi,t−1
Total BBSi,t−1
j
= αi + β j
Crediti,t
+ ui,t .
Total BBSi,t−1
(9)
j
Here BBSi,t − BBSi,t−1 refers to the one year balance sheet growth of instrument j in country i at
time t. This growth is scaled by the total size of the balance sheet in t − 1. On the right hand side, the
familiar gross inflow of capital over a one year period (Crediti,t ), is likewise scaled by total balance
sheet size. All specifications additionally include country fixed effects
Panel A of Table 6 shows in (1) that one additional dollar of gross capital inflows is, on average,
related to a balance sheet expansion of 30 cents.25 Out of these, 16 are funded via increased deposits
in (2), 9 via debt instruments (3), and the remainder with increases in reserves and other instruments
in (4) and (5). Except for the low (2 cent) increase in reserves, the results are statistically significant at
the 5 percent level. These findings align with the argument that the predominant form of short-term
foreign investment in the interwar period was deposit holdings in foreign banks (Schnabel, 2004;
Collet and Postel-Vinay, 2021). Additionally, James (1984) describes how the expansion in trade
contributed to the rise of short-term trade credits in international finance, which were increasingly
repurposed to fund domestic investments toward the end of the 1920s. The substantial amount of
long-term flows shown, e.g., in Figure 1 is less likely to appear as a bank liability as it mostly reflects
increased international trading in government securities (De Broeck et al., 2018; End et al., 2019).
While the data lacks information on the maturities of bank debt, this suggests that at least a portion
24Additionally, sudden increases in loanable funds (via e.g. capital inflows) tend to lower bank lending standards, with
assets with low credit ratings being difficult to liquidate during busts (Dell’Ariccia and Marquez, 2006)
25 This relatively low aggregate value can be explained by considering that (mostly long-term) gross inflows also include
transactions that do not pass through domestic bank balance sheets, such as FDI, land acquisitions, or purchases of
government debt by any non-bank agent. Additionally, capital inflows in the BoP cannot distinguish between capital
repatriation and investments from abroad. Consequently, inflows might even be related to reductions in domestic BBS if
banks liquidate foreign holdings, leading to capital inflows but reductions in balance sheet size. Similarly, capital outflows
do not distinguish between capital flight and foreign investment, so the relationship between capital outflows and bank
balance sheet size is not necessarily negative. For banks expanding their foreign assets, it may even be positive.
30
Table 6: Foreign capital inflows and changes in bank balance sheet size
∆Sizei,t
(1)
∆Liabilitiesi,t
Deposits
(2)
Debt
(3)
∆Assetsi,t
Reserves
(4)
Other
(5)
Loans
(6)
Bonds
(7)
Sec. ̸= Bonds
(8)
Other
(9)
Panel A: Gross capital flows and bank balance sheets
Crediti,t
0.30∗∗
(0.12)
0.16∗∗
(0.07)
0.09∗∗
(0.04)
0.02
(0.02)
0.03∗∗
(0.02)
0.13∗∗
(0.06)
0.10∗∗
(0.04)
0.02
(0.02)
0.04
(0.03)
R2
0.085
0.051
0.112
0.014
0.033
0.045
0.062
0.002
0.014
Panel B: Net capital flows and bank balance sheets
Balancei,t
0.01
(0.03)
0.00
(0.03)
0.01∗∗
(0.01)
-0.01∗∗
(0.00)
-0.00
(0.00)
0.03
(0.02)
0.00
(0.01)
-0.01∗
(0.01)
-0.01
(0.00)
R2
Country fixed effects
Observations
0.000
✓
363
0.000
✓
363
0.008
✓
363
0.005
✓
363
0.000
✓
363
0.008
✓
363
0.001
✓
363
0.008
✓
363
0.001
✓
363
Notes: This table presents ’pass-through’ coefficients from gross foreign credit (Panel A) and net foreign credit (Panel B) to changes in
domestic bank balance sheets (BBS). Bank balance sheet changes are further decomposed into the changes in deposits, debt instruments
(including cheques, drafts, acceptances and re-discounts, interbank credit), reserves and other instruments on the liability side and loans,
securities, cash and other instruments on the asset side. Changes are computed based on Equation 9. By construction, the individual
components on the asset and liability side both sum to total balance sheet growth in (1). All specifications include country fixed effects.
Standard errors in parentheses are clustered on the country level and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
of the coefficient for foreign-financed bank debt relates to short-term instruments.
To uncover a potential increase in mismatched maturities, columns (6) to (9) consider in which
types of assets banks invested their additional funds. Column (6) reveals that slightly less than half,
13 out of 30 cents, is related to an expansion in loans, while 10 cents (column (7)) are allocated to
investments in bonds. Securities (excluding bonds) and other instruments grow only marginally,
with 2 and 4 cents respectively. This allocation in assets aligns with findings from Collet and
Postel-Vinay (2021), who observe that banks increased their lending to municipal governments,
particularly through long-term municipal bonds, upon receiving foreign capital, and with End et
al. (2019), who document a rapid expansion of internationally traded long-term government debt.
Under the assumption that, in addition to short-term foreign deposits, at least part of the increase
in bank debt financed by foreign sources is short-term (as argued in James (1984)), these findings
illustrate a substantial transformation of short-term foreign financing into long-term investment,
thereby increasing financial fragility (Obstfeld, 2012).
Panel B replicates the analysis for net inflows, showing a correlation with balance sheet expansions
of virtually zero across all specifications, and confirming that surplus and deficit countries may
experience foreign-financed balance sheet growth to a similar degree (Obstfeld, 2012). This suggests
that gross flows are key to understanding the additional risk banks, exposed to foreign inflows, take
on when transforming short-term foreign liabilities into long-term (domestic) assets.
31
Table 7: Foreign credit, interest payments to foreigners and economic growth
∆2 Yi,t+2
Interest Paymentsi,t+1
Σ2j=0 Crediti,t− j
(1)
(2)
(3)
0.46∗∗∗
0.38∗∗∗
0.48∗∗∗
(0.10)
(0.09)
(0.11)
Σ2j=0 Balancei,t− j
(4)
(5)
(6)
-0.04∗∗∗
(0.01)
-0.04∗∗∗
(0.01)
-0.03∗∗∗
(0.01)
-0.00
(0.01)
-0.00
(0.01)
0.11
(0.07)
Σ2j=0 Debiti,t− j
-0.07
(0.08)
Interest Paymentsi,t
Net Interest Paymentsi,t
-0.01∗∗
(0.00)
Crediti,t
R2
Country fixed effects
Lagged Growth
Observations
0.225
✓
✓
332
0.232
✓
✓
332
0.230
✓
✓
332
0.239
✓
✓
350
0.239
✓
✓
350
0.252
✓
✓
350
Notes: This table shows in columns (1) to (3) that interest payments to foreigners in t + 1 increase in past gross foreign inflows. In columns
(4) to (6) the dependent variable is log GDP growth from t to t + 2, which is shown to be negatively related to interest payments to
foreigners in t. All specifications control for country fixed effects and two lags of GDP growth. Standard errors in parentheses are dually
clustered on country and year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
6.3.
Interest payments to foreign creditors
Borrowing today comes with interest payments in the future (Drehmann et al., 2023), which, when
borrowed from abroad, flow to foreigners. This has long been acknowledged as a drain on national
income and as a wealth transfer abroad (Lerner, 1948). More recent authors phrase a similar insight in
terms of differing marginal propensities to consume out of additional income between creditors and
debtors. In this view, domestic demand may be depressed when debtors face financial constraints
and creditors have a lower marginal propensity to spend, resulting in an aggregate demand loss
(Eggertsson and Krugman, 2012). In the case of foreign creditors, this scenario likely applies, as they
presumably have a low propensity to spend additional income in the countries of their debtors.
The Balance of Payments allows me to directly test the implication of th first part of this argument.
While gross inflows are recorded in the capital account, the current account contains information
on ”interest and dividends paid to foreigners”. Column (1) in Table 7 confirms the intuition that
interest payments to foreigners in t + 1 increase in past foreign borrowing. In columns (2) and (3)
I confirm that this is not a spurious link between two Balance of Payments variables. Controlling
for net inflows reveals that net inflows also have a slightly positive relationshpi with future interest
payments, which makes conceptual sense since net inflows, by definition, also partially capture the
gross inflow of capital. Gross capital outflows instead, as shown in column (3), are unrelated to
future interest payments to foreigners.
Columns (4) to (6) test the second part of the argument. Column (4) links the present level of
interest payments to foreigners to future GDP growth, producing a large and negatively significant
32
coefficient. This association might be partially offset if a country receives interest payments on its own
foreign lending. Column (5) therefore includes the net interest payments to foreigners, which does
not affect the coefficient for gross payments in any way. Finally, column (6) includes gross inflows at
time t to make sure the negative coefficient is not driven by a mechanical link between current and
capital account items within the BoP. Again, the result for interest payments remains unchanged. The
coefficient for credit, despite covering only one year of inflows, maintains its negative significance
but is reduced in size, suggesting that the negative link between foreign borrowing and business
cycle dynamics may indeed be partially attributed to interest payments to foreigners.
7.
Instrumental Variable Results
What are the underlying reasons behind the influx of foreign capital, and do they matter? One
possibility is that inflows are determined by domestic demand for capital, the other that they are
driven by foreign capital supply. As the baseline specification cannot distinguish between these
factors, this section proposes an instrumental variable approach to isolate foreign capital supply
from country-idiosyncratic demand for capital. In addition to exploiting exogenous variation in
capital inflows, this distinction also addresses a potential downward bias in the baseline. Borrowing
against future fundamentals or seeking external funds to finance investment is unlikely to have
adverse aggregate effects, while supply-induced inflows are unrelated to the domestic economy
and likely drive the negative association between foreign inflows and economic downturns (Rey,
2013; Miranda-Agrippino and Rey, 2020). An instrument that succeeds in identifying foreign capital
supply should consequently lead to an upward correction of the OLS coefficient.
7.1.
The United States as a creditor nation
The instrumental variable approach exploits the special role of the United States, which had become
the world’s greatest creditor nation during the boom years of the Roaring Twenties.26 This fact, well
appreciated by contemporaries, resulted in a magnitude of publications addressing the American role
in world finance, and trying to assess how large the American investment position abroad actually
was.27 Later scholars observed the change from ’debtor to creditor nation’ as America becoming the
’world’s banker’ (Woodruff, 1975). When the US drastically reduced its foreign lending during the
Great Depression, countries that were previously subject to most inflows from the US, consequently
now experienced the most severe reduction in available foreign capital. To test this line of reasoning, I
collect data from Dickens (1931, 1930) and Lewis and Schlotterbeck (1938), and construct the bilateral
portfolio investment position of the United States with individual countries.
26When exactly the US overtook the UK as the leading financial center is subject to debate, but Eichengreen and Flandreau
(2009) and Eichengreen and Flandreau (2012) suggest that it likely occurred well before the Great Depression.
27 See Dickens (1931, 1930); Lewis and Schlotterbeck (1938) for data collection and Jolliffe (1935) for a narrative approach.
33
To highlight the significance of US capital for domestic financial systems, I again compute
pass-through coefficients to domestic bank balance sheets, using the method from section 6.2. The
results in Table A8.20 indicate that one additional dollar of US portfolio investment was, on average,
associated with a balance sheet expansion of $0.97.28 In line with previous results, the majority
($0.68) of this expansion came in the form of foreign deposits. On the asset side, $0.45 was lent as
domestic credit, while $0.28 was used to finance bond purchases. This confirms a direct connection
between US portfolio investments and the financial systems of recipient countries, with most US
funds providing liquidity to credit and financial markets. To express the importance of US capital in
relative terms, Figure A8.10 plots the ratio of the US portfolio position to bank balance sheet size
for Germany, the UK, France, and Japan. Although magnitudes and pre-crisis trends vary, this ratio
sharply declines in all four countries from 1931/32 onward, indicating that the US portfolio position
contracted more rapidly than domestic bank balance sheets during that period.
Having established a link from US foreign investments to domestic financial conditions, I now
turn to the economic relevance of having strong financial ties to the United States. Concretely, I
regress three year future GDP growth on the lagged ratio of US portfolio investments to domestic
bank balance sheet size for every year between 1923 and 1934 individually. The results in Figure A8.11
show that closer financial ties to the US during the expansionary years before and after the Great
Depression are positively associated with domestic GDP growth. This relationship, however, turns
significantly negative towards the end of the 1920s and the onset of the Great Depression, showing
that exposure to potential US capital withdrawals had adverse effects on the economy.
7.2.
Instrumenting gross foreign inflows
Arguably, the flight of US capital from (or supply to) any individual country is driven as much by that
country’s economic conditions as by the United States’ willingness to invest abroad. The aggregate
withdrawal (or expansion) of foreign investments by the United States, however, is more likely to be
driven by decisions taken in the US rather than their individual partner countries. To exploit this
variation in foreign inflows not driven by specific countries’ economic conditions, but by changing
US conditions, I construct an instrumental variable for capital inflows based on bilateral exposure to
the US. Specifically, I interact exposure to US portfolio investments in 1926 with the change in the
total US portfolio position. To minimize potential endogeneity with domestic economic conditions,
the change in the total US portfolio position is computed excluding changes in investments to the
instrumented country i.29 Equation 10 shows the construction of the instrument.
28 This value is significantly higher than the $0.3 pass-through shown for total gross inflows. This has two reasons: First,
gross inflows include all cross-border capital flows and thus a higher share of transactions that do not enter domestic bank
liabilities (FDI, land purchases, government bonds, inter-firm trade credit) compared to portfolio investments. Second,
gross inflows do not distinguish between capital repatriation and investments from abroad, and, consequently, also contain
inflows likely to reduce bank balance sheet size (e.g. banks repatriating their foreign asset holdings).
29While the construction of this shift-share instrument follows Bartik (1991), it is not an original Bartik instrument where
exposures (the shares) add up to one and the shock (the shift) is identical for all. Here, the shares are a country-specific
34
Interaction IVi,t =
USPi,1926
× ∆USPj,t ,
BBSi,1926 ∑
j ̸ =i
(10)
where USPi,1926 is the United States’ portfolio position in country i in 1926, scaled by that countries
bank balance sheet size BBSi,1926 .30 The exposure measure is time invariant which leaves the second
term to create variation in the instrument. This second term is the total change in the US portfolio
position, excluding the instrumented country i to make sure that US capital supply is not conflated
with capital demand by country i. Finally, I normalize the instrument to make the reduced form
coefficients comparable to the coefficients obtained from Balance of Payments variables.
The first stage of this instrument is shown in a scatterplot against gross foreign inflows in
Figure A8.12. The relationship between the variables is, in line with intuition, strongly positive and
strengthened upon including control variables and country fixed effects in the second panel. Table 8
reports OLS coefficients along with reduced form and instrumented estimates for increasing forecast
horizons of GDP, mirroring the baseline specification. Differing from the baseline, the independent
variables are reduced to yearly capital inflows to match the instrument’s time horizon. Restricting
the sample to observations where both the instrument (post-1926 and excluding the US) and the
BoP variables are available reduces the total number of observations to slightly above 200, yet gross
yearly inflows remain robustly associated with lower future GDP growth. The reduced form and
instrumented coefficients are likewise highly significant but tend to be larger than the OLS baseline.
The Kleibergen-Paap statistic of around 20 further confirms the visual impression of a good first-stage
fit. These findings suggest a baseline bias towards zero when domestic demand and foreign capital
supply cannot be distinguished, and, more generally, attribute an important role to supply-driven
gross inflows in the relationship between capital flows and economic outcomes.
Table A8.21 in the appendix extends the instrument’s application to predicting financial crises, reporting OLS, reduced form, and instrumented coefficients individually. Again, all three specifications
produce highly significant results (post-1926 sample, excluding the US), with instrumented coefficients again being larger than the baseline. Instrumenting capital inflows with a measure for foreign
capital supply also enables me to effectively shut down the channel of capital inflows being driven by
a bank’s decision for capital repatriation, allowing cleaner identification of foreign-financed domestic
balance sheet growth. Table A8.22 applies the instrument to the balance sheet decomposition exercise
from section 6.2. In line with intuition, coefficient size doubles compared to OLS results estimated
on the same sample, with especially domestic lending taking an outsized role in asset expansion.
variable, as it matters more how important US capital was for a given country’s financial system, rather than how much of
all US investments it received. Similarly, the shift is not identical for all but varies across countries, as each country’s part
in US foreign investments is excluded to reduce endogeneity. In both cases, however, the key identifying assumption is
that the pre-existing exposure measure is exogenous to the aggregate developments (Goldsmith-Pinkham et al., 2020).
301926 is chosen as it is the first year for which comprehensive data is available for many countries. The specification
provides similar estimates when choosing earlier years, which reduces the sample along the cross-sectional dimension, as
well as later years, which reduces the sample along the time dimension.
35
Table 8: Gross inflows and GDP dynamics, instrumental variable results
∆2 Yi,t+2
OLS
(1)
Crediti,t
-0.03∗∗
(0.01)
∆3 Yi,t+3
IV
(3)
OLS
(4)
-0.05∗∗
(0.02)
-0.04∗∗∗
(0.01)
-0.02∗
(0.01)
Interaction IVi,t
Country fixed effects
Lagged Growth
Kleibergen-Paap Weak ID
Observations
Reduced
(2)
✓
✓
✓
✓
225
225
Reduced
(5)
∆4 Yi,t+4
IV
(6)
OLS
(7)
-0.09∗∗∗
(0.03)
-0.04∗∗∗
(0.01)
-0.05∗∗∗
(0.02)
✓
✓
27.25
225
✓
✓
✓
✓
225
225
Reduced
(8)
IV
(9)
-0.13∗∗∗
(0.03)
-0.06∗∗∗
(0.02)
✓
✓
28.30
225
✓
✓
✓
✓
216
216
✓
✓
18.25
216
Notes: This table presents OLS, reduced form, and instrumented coefficients for a regression of log GDP growth between t and t + h on
gross inflows at time t. The instrument is constructed as described in Equation 10 and used to instrument gross inflows in columns (3), (6)
and (9). All specifications control for country fixed effects, a two year distributed lag of GDP growth and net capital flows. Standard
errors in parentheses are dually clustered on country and year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
Together, these findings confirm that supply-driven inflows played an important role in increasing
financial fragility and crisis likelihood.
Robustness: Fixing the exposure to 1926 reduces data coverage, and does not allow for dynamically adjusting exposures over time. For robustness, I re-estimate the previous table with exposure
alternatively defined as a moving average of the previous two years, instead of 1926. While this
allows for the inclusion of data points before 1926 wherever available, it opens up the possibility of
variation in the instrument being driven by changes in the exposure measure. Being aware of this
trade-off, I report this specification in Table A8.23, confirming the previous results in a slightly larger
sample. I address further concerns about the instrument’s construction by building an instrument
that fully adheres to the formal requirements for a Bartik-instrument. This version normalizes
exposure shares to sum to one, and interacts them with the total shift in US portfolio investments,
identical across all countries. Results in Table A8.24 only marginally differ from the baseline.
A potential caveat of the interaction instrument is that it might be too US-centric, given that the
US had only just overtaken the U.K. as the primary creditor country, with the UK remaining a close
second (Ritschl, 2002b). To model global capital supply in a more general way, I follow Aldasoro
et al. (2020), who construct a measure of the Global Financial Cycle using principal component
analysis.31 To minimize the endogeneity of the global cycle to any individual country, I estimate it
for every country individually as the first principal component of capital inflows over all countries,
excluding country i itself. I then calculate normalized three-year sums of the GFC-measure, identical
to the baseline cumulative BoP variables, and establish the relevance of the instrument in a firststage scatterplot in Figure A8.13, showing a strong positive correlation. Finally, I instrument gross
foreign inflows with the GFC-measure in Table A8.25. Again, the reduced form and especially
the instrumented coefficients turn out to be larger than the baseline OLS coefficients and highly
significant across all horizons, reaffirming the intuition of a baseline bias towards zero.
31 The concept of the GFC, pioneered by Rey (2013), describes the spillover of global financial conditions into domestic
economies, irrespective of the domestic financial cycle. See also: Miranda-Agrippino and Rey (2020); Aldasoro et al. (2020).
36
8.
Conclusion
The Great Depression, as the pivotal event of the interwar years, has been studied by innumerable
scholars. The classical perspective on what preceded it, how it played out, and what followed it,
however, seems to be remarkably consensual (James, 1992). This paper shows that the established
approach from the perspective of international finance, centering on the Gold Standard and net
capital flows, can be recreated with novel data from the League of Nations, but also that this approach
needs to be expanded. Telling the story of the boom-bust dynamics around the Great Depression
requires discussing the boom in gross international credit that went bust.
Gross foreign inflows are decisive for adverse future economic outcomes. They link international
capital flows to economic downturns and financial crises, and are an important factor in the severity
of post-crisis recessions. This relationship goes largely unnoticed when focusing on net capital flows,
which are unable to fully capture exposure to global capital movements and the buildup of foreign
debt positions, echoing the mixed success net measures have had in the literature. These results
hold across a battery of robustness checks and can be rationalized by foreign debt service payments,
increased fragility in domestic financial systems, and the dynamics of the Gold Standard.
Together, my findings can be considered a variation of the Golden Fetters argument (Eichengreen,
1992), as the Gold Standard was instrumental in exposing countries to the documented dynamics
by increasing their integration into the global financial system. Simultaneously, it deprived them
of the tools to counter the adverse effects of foreign credit. Not being on the Gold Standard
consequently provided a margin of adjustment that helped to partially shield countries from the
harmful consequences of gross foreign credit.
37
References
Accominotti, Olivier and Barry Eichengreen, “The mother of all sudden stops: capital flows and reversals in
europe, 1919–32,” The Economic History Review, 2016, 69 (2), 469–492.
Adalet, Muge and Barry Eichengreen, “Current account reversals: always a problem?,” in “G7 Current
Account Imbalances: Sustainability and Adjustment,” University of Chicago Press, 2007, pp. 205–246.
Adrian, Tobias and Hyun Song Shin, “Liquidity and leverage,” Journal of financial intermediation, 2010, 19 (3),
418–437.
Albers, Thilo Nils Hendrik, “The prelude and global impact of the Great Depression: evidence from a new
macroeconomic dataset,” Explorations in Economic History, 2018, 70, 150–163.
Aldasoro, Iñaki, Stefan Avdjiev, Claudio EV Borio, and Piti Disyatat, “Global and domestic financial cycles:
variations on a theme,” BIS Working Papers, 2020, 864.
Atkeson, Andrew, Jonathan Heathcote, and Fabrizio Perri, “The end of privilege: A reexamination of the net
foreign asset position of the United States,” 2022.
Baron, Matthew, Emil Verner, and Wei Xiong, “Banking Crises Without Panics,” The Quarterly Journal of
Economics, 2021, 136 (1), 51–113.
Bartik, Timothy J, “Who benefits from state and local economic development policies?,” 1991.
Baubeau, Patrice, Eric Monnet, Angelo Riva, and Stefano Ungaro, “Flight-to-safety and the credit crunch: a
new history of the banking crises in France during the Great Depression,” The Economic History Review, 2021,
74 (1), 223–250.
Bernanke, Ben S, “The macroeconomics of the Great Depression: A comparative approach,” 1994.
Bernanke, Ben S., “The Global Saving Glut and the US Current Account Deficit,” Board of Governors of the
Federal Reserve System (U.S.), 2005, (77).
Bernanke, Ben S, “Essays on the great depression,” in “in,” Princeton University Press, 2009.
Bernanke, Ben S and Harold James, “The gold standard, deflation, and financial crisis in the Great Depression:
An international comparison,” 1990.
Bleaney, Michael and Mo Tian, “Net foreign assets, real exchange rates and net exports revisited,” Technical
Report, CREDIT Research Paper 2013.
Bolt, Jutta and Jan Luiten van Zanden, Maddison style estimates of the evolution of the world economy. A new 2020
update, Maddison Project Database, version 2020, 2020.
Bordo, Michael D and Finn E Kydland, “The gold standard as a rule,” in “Gold Standard In Theory and
History,” Routledge, 2005, pp. 77–99.
Bordo, Michael D and Michael Edelstein, “Was Adherence to the Gold Standard a” Good Housekeeping Seal
of Approval” During the Interwar Period?,” 1999.
Bordo, Michael D, Barry Eichengreen, Daniela Klingebiel, and Maria Soledad Martinez-Peria, “Is the crisis
problem growing more severe?,” Economic policy, 2001, 16 (32), 52–82.
Borio, Claudio, “On the centrality of the current account in international economics,” Journal of International
Money and Finance, 2016, 68, 266–274.
Borio, Claudio and Piti Disyatat, “Capital flows and the current account: Taking financing (more) seriously,”
2015.
Borio, Claudio, Harold James, and Hyun Song Shin, “The International Monetary and Financial System: a
Capital Account Historical Perspective,” BIS Working Paper, 2014, 457.
Broner, Fernando, Tatiana Didier, Aitor Erce, and Sergio L. Schmukler, “Gross Capital Flows: Dynamics and
Crises,” Journal of Monetary Economics, 2013, 60 (1), 113–133.
Caballero, Julián A., “Do Surges in International Capital Inflows Influence the Likelihood of Banking Crises?,”
The Economic Journal, 2016, 126 (591), 281–316.
Caballero, Ricardo J. and Alp Simsek, “A Model of Fickle Capital Flows and Retrenchment,” Journal of Political
38
Economy, 2020, 128 (6), 2288–2328.
Calderon, Cesar and Megumi Kubota, “Gross inflows gone wild: gross capital inflows, credit booms and
crises,” World Bank policy research working paper, 2012, (6270).
Calvo, Guillermo A, “Capital flows and capital-market crises: the simple economics of sudden stops,” Journal
of applied Economics, 1998, 1 (1), 35–54.
Chodorow-Reich, Gabriel, “The employment effects of credit market disruptions: Firm-level evidence from
the 2008–9 financial crisis,” The Quarterly Journal of Economics, 2014, 129 (1), 1–59.
Collet, Stéphanie and Natacha Postel-Vinay, “Hot money inflows and bank risk-taking: Germany from the
1920s to the great depression,” 2021.
Crowther, Geoffrey, An outline of money, T. Nelson, 1949.
Da Rocha, Bruno T and Solomos Solomou, “The effects of systemic banking crises in the inter-war period,”
Journal of International Money and Finance, 2015, 54, 35–49.
Davis, J Scott, Adrienne Mack, Wesley Phoa, and Anne Vandenabeele, “Credit booms, banking crises, and
the current account,” Journal of International Money and Finance, 2016, 60, 360–377.
De Broeck, Mark, Era Dabla-Norris, Nicolas End, and Marina Marinkov, “The Debt Web,” Finance and
Development, 2018, 55 (001).
Degorce, Victor and Eric Monnet, “The great depression as a saving glut,” 2020.
Dell’Ariccia, Giovanni and Robert Marquez, “Lending booms and lending standards,” The journal of finance,
2006, 61 (5), 2511–2546.
Deutsche Bundesbank, Währung und Wirtschaft in Deutschland 1876-1975, Fritz Knapp GmbH, 1976.
Dickens, Paul DeWitt, American Direct Investments in Foreign Countries–1930 number 1, US Government
Printing Office, 1930.
Dickens, Paul DeWitt, A new estimate of American investments abroad number 767, US Government Printing
Office, 1931.
Diebold, Lukas and Björn Richter, “When two become one: Foreign capital and household credit expansion,”
Available at SSRN 3817595, 2021.
Drehmann, Mathias, Mikael Juselius, and Anton Korinek, “Long-term debt propagation and real reversals,”
BIS Working Papers, May 2023, (1098).
Edwards, Sebastian, “Does the current account matter?,” in “Preventing currency crises in emerging markets,”
University of Chicago Press, 2002, pp. 21–76.
Eggertsson, Gauti B and Paul Krugman, “Debt, deleveraging, and the liquidity trap: A Fisher-Minsky-Koo
approach,” The Quarterly Journal of Economics, 2012, 127 (3), 1469–1513.
Eichengreen, Barry, Golden fetters: the gold standard and the Great Depression, 1919-1939, Oxford University Press,
1992.
Eichengreen, Barry, “Globalizing capital,” in “Globalizing Capital,” Princeton University Press, 2008.
Eichengreen, Barry and Douglas A Irwin, “The slide to protectionism in the great depression: who succumbed
and why?,” The Journal of Economic History, 2010, 70 (4), 871–897.
Eichengreen, Barry and Kris James Mitchener, “The Great Depression as a credit boom gone wrong,” 2003.
Eichengreen, Barry and Marc Flandreau, “The rise and fall of the dollar (or when did the dollar replace
sterling as the leading reserve currency?),” European Review of Economic History, 2009, 13 (3), 377–411.
Eichengreen, Barry and Marc Flandreau, “The Federal Reserve, the Bank of England, and the rise of the dollar
as an international currency, 1914–1939,” Open Economies Review, 2012, 23 (1), 57–87.
Eichengreen, Barry and Peter Temin, “The gold standard and the great depression,” Contemporary European
History, 2000, 9 (2), 183–207.
Ellison, Martin, Sang Seok Lee, and Kevin Hjortshøj O’Rourke, “The Ends of 27 Big Depressions,” American
Economic Review, 2023, , Forthcoming.
39
End, Nicolas, Marina Marinkov, and Fedor Miryugin, Instruments of debtstruction: A new database of interwar
debt, International Monetary Fund, 2019.
Erce, Aitor and Enrico Mallucci, “Selective sovereign defaults,” FRB International Finance Discussion Paper,
2018, (1239).
Eren, Egemen, Andreas Schrimpf, and Fan Dora Xia, “The demand for government debt,” Available at SSRN
4466154, 2023.
Fang, Xiang, Bryan Hardy, and Karen K Lewis, “Who holds sovereign debt and why it matters,” Technical
Report, National Bureau of Economic Research 2022.
Feinstein, Charles H and Katherine Watson, “Private international capital flows in Europe in the inter-war
period,” Banking, currency and finance in Europe between the wars, 1995, pp. 94–130.
Fisher, Irving, “Are booms and depressions transmitted internationally through monetary standards?,” Bulletin
de l’Institut International de Statistique, 1935, 28, 1–29.
Forbes, Kristin J. and Francis E. Warnock, “Capital flow waves: Surges, Stops, Flight, and Retrenchment,”
Journal of International Economics, 2012, 88 (2), 235–251. NBER Global.
Goldsmith-Pinkham, Paul, Isaac Sorkin, and Henry Swift, “Bartik instruments: What, when, why, and how,”
American Economic Review, 2020, 110 (8), 2586–2624.
Gourinchas, Pierre-Olivier and Maurice Obstfeld, “Stories of the Twentieth Century for the Twenty-First,”
American Economic Journal: Macroeconomics, 2012, 4 (1), 226–65.
Grossman, Richard S, “The shoe that didn’t drop: explaining banking stability during the Great Depression,”
The Journal of Economic History, 1994, 54 (3), 654–682.
Guerrieri, Veronica and Guido Lorenzoni, “Credit crises, precautionary savings, and the liquidity trap,” The
Quarterly Journal of Economics, 2017, 132 (3), 1427–1467.
Hoffmann, Mathias and Ulrich Woitek, “Emerging from the War: Current Accounts and the International
Business Cycle 1885-1939,” 2010.
Hume, David, Essays moral, political, and literary, Vol. 1, Longmans, Green, 1758.
IMF, Balance of Payments and International Investment Position Manual, Washington: International Monetary
Fund, Publication Services, 2009.
Irwin, Douglas A, “The French gold sink and the great deflation of 1929-32,” Cato Papers, 2012, 2, 1–52.
Irwin, Douglas A, “Who anticipated the great depression? Gustav Cassel versus Keynes and Hayek on the
interwar gold standard,” Journal of Money, Credit and Banking, 2014, 46 (1), 199–227.
James, Harold, “The causes of the German banking crisis of 1931,” Economic History Review, 1984, pp. 68–87.
James, Harold, “Financial flows across frontiers during the interwar depression,” Economic History Review,
1992, pp. 594–613.
Jolliffe, Muriel Florence, “The United States as a financial centre, 1919-1933: with reference to imports and
exports of capital,” (No Title), 1935.
Jordà, Òscar, “Estimation and Inference of Impulse Responses by Local Projections,” American Economic Review,
2005, 95 (1), 161–182.
Jordà, Òscar, Moritz Schularick, and Alan M. Taylor, “Financial Crises, Credit Booms, and External Imbalances: 140 Years of Lessons,” IMF Economic Review, 2011, 59 (2), 340–378.
Jordà, Òscar, Moritz Schularick, and Alan M. Taylor, “When Credit Bites Back,” Journal of Money, Credit and
Banking, 2013, 45 (s2), 3–28.
Keynes, John Maynard, The General Theory of Employment, Interest and Money, MacMillian, 1936.
Keynes, John Maynard, “Post-war currency policy,” 1980): The collected writings on John Maynard Keynes, 1941,
25, 1940–1944.
Kiley, Michael T., “What Macroeconomic Conditions lead Financial Crises?,” Journal of International Money and
Finance, 2021, 111, 102316.
40
Kindleberger, Charles P., Manias, Panics and Crashes: A History of Financial Crises, Palgrave Macmillan, 1978.
Klimantas, Adomas and Aras Zirgulis, “A new estimate of Lithuanian GDP for 1937: How does interwar
Lithuania compare?,” Cliometrica, 2020, 14 (2), 227–281.
Krishnamurthy, Arvind and Tyler Muir, “How Credit Cycles Across a Financial Crisis,” NBER Working Paper,
2017, (23850).
Lane, Philip R and Gian Maria Milesi-Ferretti, “The transfer problem revisited: Net foreign assets and real
exchange rates,” Review of Economics and Statistics, 2004, 86 (4), 841–857.
League of Nations, (LoN), Statistical yearbook of the League of Nations, Economic and Financial Section, 1921-1939.
League of Nations, (LoN), Memorandum on Interantional Trade and Balance of Payments, Economic Intelligence
Service, 1930-1932.
League of Nations, (LoN), Memorandum on Commercial Banks, Economic Intelligence Service, 1931-1940.
League of Nations, (LoN), Balance of Payments, Economic Intelligence Service, 1933-1939.
League of Nations, (LoN), Money and Banking: Monetary Review, Economic Intelligence Service, 1936-1939.
Lerner, Abba P, “The burden of the national debt,” Income, Employment and Public Policy: Essays in Honor of
Alvin Hansen, 1948, pp. 255–275.
Lewis, Cleona and Karl T Schlotterbeck, America’s stake in international investments number 75, Brookings
Institution, 1938.
Liadze, Iana, Ray Barrell, and Philip E. Davis, “The Impact of Global Imbalances: Does the Current Account
Balance help to Predict Banking Crises in OECD Countries?,” NIESR Discussion Papers, 2010, (351).
Magud, Nicolas E, Carmen M Reinhart, and Esteban R Vesperoni, “Capital inflows, exchange rate flexibility
and credit booms,” Review of Development Economics, 2014, 18 (3), 415–430.
Mian, Atif, Amir Sufi, and Emil Verner, “How Does Credit Supply Expansion Affect the Real Economy? The
Productive Capacity and Household Demand Channels,” The Journal of Finance, 2020, 75 (2), 949–994.
Mian, Atif R., Amir Sufi, and Emil Verner, “Household Debt and Business Cycles Worldwide,” Quarterly
Journal of Economics, 2017, 132 (4), 1755–1817.
Milesi-Ferretti, Gian-Maria and Cédric Tille, “The great retrenchment: international capital flows during the
global financial crisis,” Economic policy, 2011, 26 (66), 289–346.
Miranda-Agrippino, Silvia and Hélène Rey, “U.S. Monetary Policy and the Global Financial Cycle,” The
Review of Economic Studies, 05 2020, 87 (6), 2754–2776.
Mitchener, Kris James and Kirsten Wandschneider, “Capital controls and recovery from the financial crisis of
the 1930s,” Journal of International Economics, 2015, 95 (2), 188–201.
Müller, Karsten and Emil Verner, “Credit allocation and macroeconomic fluctuations,” Review of Economic
Studies, 2024, 91 (6), 3645–3676.
Norkus, Zenonas and Jurgita Markevičiūtė, “New estimation of the gross domestic product in Baltic countries
in 1913–1938,” Cliometrica, 2021, 15 (3), 565–674.
Obstfeld, Maurice, “Does the Current Account Still Matter?,” American Economic Review, 2012, 102 (3), 1–23.
Obstfeld, Maurice et al., “Expanding gross asset positions and the international monetary system,” in
“Presentation at the Symposium “Macroeconomic Challenge: The Decade Ahead”, Jackson Hole” 2010.
Obstfeld, Maurice, Jay C Shambaugh, and Alan M Taylor, “Monetary sovereignty, exchange rates, and capital
controls: The trilemma in the interwar period,” IMf staff papers, 2004, 51 (Suppl 1), 75–108.
Pineau, Georges, Ettore Dorrucci, Fabio Comelli, and Angelika Lagerblom, “The accumulation of foreign
reserves,” ECB Occasional Paper, 2006, (43).
Quinn, Dennis P, “Capital account liberalization and financial globalization, 1890–1999: a synoptic view,”
International Journal of Finance and Economics, 2003, 8 (3), 189–204.
Reinhart, Carmen M. and Kenneth S. Rogoff, This Time is Different: Eight Centuries of Financial Folly, Princeton
University Press, 2009.
41
Rey, Hélène, “Dilemma not Trilemma: The Global Cycle and Monetary Policy Independence,” Proceedings Economic Policy Symposium - Jackson Hole, 2013.
Ritschl, Albrecht, “Peter Temin and the onset of the Great Depression in Germany: a reappraisal,” University
of Zürich, mimeographed, 1999.
Ritschl, Albrecht, Deutschlands Krise und Konjunktur 1924–1934, Akademie Verlag, 2002.
Ritschl, Albrecht, International capital movements and the onset of the great depression: some international evidence,
Oldenbourg Wissenschaftsverlag, 2002.
Ritschl, Albrecht, “The German transfer problem, 1920–33: a sovereign-debt perspective,” European Review of
History: Revue europeenne d’histoire, 2012, 19 (6), 943–964.
Ritschl, Albrecht and Tai-kuang Ho, “Seniority Reversals and Endogenous Sudden Stops: Some Transfer
Problem Dynamics,” 2023.
Romer, Christina D. and David H. Romer, “New Evidence on the Aftermath of Financial Crises in Advanced
Countries,” American Economic Review, 2017, 107 (10), 3072–3118.
Schmitt-Grohé, Stephanie and Martı́n Uribe, “Downward Nominal Wage Rigidity, Currency Pegs, and
Involuntary Unemployment,” Journal of Political Economy, 2016, 124 (5), 1466–1514.
Schnabel, Isabel, “The German twin crisis of 1931,” The Journal of Economic History, 2004, 64 (3), 822–871.
Schularick, Moritz and Alan M. Taylor, “Credit Booms Gone Bust: Monetary Policy, Leverage Cycles, and
Financial Crises, 1870-2008,” American Economic Review, 2012, 102 (2), 1029–61.
Shin, Hyun Song, “Global Banking Glut and Loan Risk Premium,” IMF Economic Review, 2012, 60 (2), 155–192.
Spoerer, Mark and Jochen Streb, Neue deutsche Wirtschaftsgeschichte des 20. Jahrhunderts, Oldenbourg Wissenschaftsverlag, 2013.
Tomz, Michael and Mark LJ Wright, “Do countries default in “bad times”?,” Journal of the European Economic
association, 2007, 5 (2-3), 352–360.
Wandschneider, Kirsten, “The stability of the interwar gold exchange standard: Did politics matter?,” The
Journal of Economic History, 2008, 68 (1), 151–181.
Woodruff, William, America’s Impact on the World: A Study of the Role of the United States in the World Economy,
1750-1970, Springer, 1975.
42
Online Appendix
Golden Fetters or Credit Boom Gone Bust?
A Reassessment of Capital Flows in the Interwar Period
A.
A1.
Appendix
BoP Data
Figure A1.1: Balance of Payments original data example, United States
Notes: This figure shows a snapshot of a typical Balance of Payments table for the United States, newly digitized for this paper. It is taken
from the 1934 publication of the League of Nations (1933-1939a). Coverage and quality of data may differ across countries and time.
Figure A1.2: Current account composition for the U.S., Germany, the U.K., and Japan
1924 1927 1930 1933 1936
1924 1927 1930 1933 1936
Japan
In Billion USD$
-2 -1.5 -1 -.5 0 .5 1 1.5 2
United Kingdom
In Billion USD$
-8 -6 -4 -2 0 2 4 6 8
Germany
In Billion USD$
-4 -3 -2 -1 0 1 2 3 4
In Billion USD$
-8 -6 -4 -2 0 2 4 6 8
United States
1924 1927 1930 1933 1936
Merchandise
Gold
Balance
Interest Dividends
Service / Other
1924 1927 1930 1933 1936
Notes: This figure shows the annual gross financial flows from the current account side of the balance of payments for the United States,
Germany, the United Kingdom and Japan. Figures are in billion US dollars (note that scales are country specific). Purple, green, yellow
and gray represent flows in goods, secondary incomes (interest and dividends), gold and services respectively. Black is the current account
balance.
A1
A2.
Data Documentation
A potential concern with the League of Nations data, acknowledged by the League itself, is that its
coverage regarding details on the gross inflow and outflow of capital is severely limited for some
countries, potentially introducing a large degree of data heterogeneity into a cross-country sample.
Previous economic historians, including James (1992) and Feinstein and Watson (1995), have agreed
with the League’s assessment and considered the provided information unsatisfactory.32
Specifically, these limitations manifest in incomplete ’direct’ knowledge about the precise composition of gross capital flows. In some cases, only the well-defined entries are reported under the
heading ’known capital items.’ In others, potentially large ’undefinable’ items, or ’other’ movements
of capital are included, providing more data points but introducing uncertainty about what precisely
the data contains. The League considered its own estimates of the net movement of capital to be
more reliable than its estimates of gross flows, as it could be computed via both the current and
capital accounts. Consequently, some countries resorted to ’indirectly’ estimating their gross capital
movements by introducing balancing items to their gross capital flows, perfectly equating the capital
account balance with the inverse of the current account balance.
The assumed perfect equality between account balances is unlikely to hold exactly without ex-post
balancing (due to statistical inaccuracies, revaluations, currency adjustments, and delayed reporting
practices), and the larger the difference between balances, the less accurate the reported gross flows.
To systematically evaluate the data quality, I first focus on observations where the current account
balance is unequal to the inverse capital account balance (i.e., where the two have not been made
to fit by construction). In the left panel of Figure A2.3, I consequently plot the balances against
each other only when their raw values in the LoN data are unequal. The solid line represents the
fitted relationship, while the dashed line represents perfect equality. As expected, the relationship is
negative, with frequent but apparently non-systematic deviations from absolute parity. Systematic
deviations would be indicated by the fitted line being steeper (underreporting gross outflows) or less
steep (underreporting gross inflows) than the line of equality.33
To judge how large these deviations are relative to the total of recorded capital flows, I compute
the ratio of the absolute difference between the two balances to total gross flows and plot it against
log total gross flows in the right panel. This shows, on the one hand, that deviations between the
current and capital account balance can indeed be relatively large, being close to 60% of gross flows in
the highest bin, confirming the inaccuracies in the reporting of gross flows noted by previous authors.
On the other hand, for the vast majority of data, the deviation is between 0 and 30% and is decreasing
in gross flows, which may point towards a decreasing relevance of net deviations when gross flows
32 I thank an anonymous referee for bringing these previous concerns to my attention.
33 E.g.:, If gross inflows were frequently unknown or underreported, the capital account balance would be relatively
smaller, as it is defined as credits minus debits. Large negative current account entries would often not be mirrored by
correspondingly large positive capital account entries, causing the fitted line to be less steep than the line of equality.
A2
3
ARG1930
CAN1930
IND1926
HUN1928
HUN1927
JPN1926
USA1935
NOR1923
DNK1931
FIN1931
.45
FIN1928
.15
0
-1.5
.3
ROU1929
TUR1926
SWE1923
USA1936
THA1937
NLD1932
CZE1932 IND1927
ALB1931
CZE1931
ARG1921 NOR1924
JPN1925
TUR1927
CAN1929
AUS1929
THA1935 IRQ1933
DNK1923
SWE1931
NOR1925
IND1930
NZL1934
NOR1930
NLD1931
NLD1929
NLD1930
JPN1928
ALB1932
DNK1926
SWE1926
DNK1927
CZE1933
NOR1928
ALB1930
HUN1931
ARG1926
NLD1931
IND1924
SWE1924
ARG1928
FIN1930 THA1936
USA1934
IND1928
AUS1928
NZL1930
NLD1933
FIN1929
IRQ1932JPN1927
USA1927 THA1933
USA1923
POL1924
ZAF1934
SWE1935
NOR1931
ARG1924
ARG1923
SWE1936
DNK1934
NLD1926
HUN1930
HUN1929
NOR1927
CZE1934
TUR1929
IRQ1930
CAN1931
CZE1935
IND1929
CZE1936
CZE1925 ARG1927 NZL1927
DNK1924
DNK1930 USA1929
ARG1936
CZE1937
ARG1933
USA1926
HUN1926
JPN1934
ROU1928
USA1922
NOR1929
IND1937
IRL1934
ALB1933USA1932
JPN1929
POL1937
DNK1935
SWE1932
FIN1927
CAN1932
CAN1927
IRL1933
CZE1930
FIN1926
NLD1933
USA1933
AUS1934
ZAF1937
CAN1933
NZL1937
DNK1936
HUN1924
NLD1928
THA1934
TUR1928
NLD1936
DNK1925
ARG1922
NOR1926
NLD1934
JPN1936
ARG1925
NZL1936
SWE1927
NLD1935
GRC1936
JPN1933
USA1931
SWE1929
NZL1931
FIN1935
ARG1934
IND1923
CZE1929
USA1924
JPN1935
POL1932
IRL1937 CAN1934 ROU1926
POL1931
IND1932 NZL1932
IND1935
GRC1933
ARG1935
SWE1925
POL1933
ROU1927
NLD1934
NLD1937
AUS1930
SWE1933
CAN1926
USA1925
CAN1928
POL1935
USA1930
FIN1932
POL1934
ZAF1933
HUN1923
FIN1937
HUN1932
NOR1936
IND1934
FIN1936
AUS1935
HUN1933
ZAF1935
FIN1934
ARG1929
CAN1935
DNK1929
GRC1934
NZL1935
NLD1927
JPN1932
NLD1932 SWE1928
HUN1934
CAN1937
USA1921
DNK1933
ARG1932
NLD1936
NLD1935
USA1928
IRQ1934
IRL1935
SWE1934
HUN1935
IND1925
HUN1936
NOR1934
NOR1933
CAN1936
NOR1932
ALB1927
IND1936
IND1933
IRQ1931
AUS1931
ZAF1936
IND1931
DNK1937
CAN1925
ALB1926
AUS1933
IRL1931
AUS1932
CZE1926
AUS1936
IRL1936
DNK1932
JPN1930
CZE1928
JPN1931
NOR1937
1.5
ARG1937
CZE1927
NLD1930
FIN1933
NLD1925
SWE1930
0
Capital Account Balance
Deviation of Balances
.6
Account Balances
Balance Deviation / Total Gross Flows
Figure A2.3: Data validation, capital and current account balance deviations
-3
NZL1933
-3
-1.5
0
1.5
Current Account Balance
3
14
16
18
20
Log Gross Capital Flows
22
Notes: This figure shows in the left panel a scatterplot of the current against the capital account balance, only for observations where their
inverses are unequal in the raw data of the League of Nations. The dashed line represents perfect (inverse) equality between the two, the
solid line is the fitted linear relationship. The right panel plots the absolute deviations between the two balances relative to gross flows
against total log gross flows in a binned scatterplot. For further explanations see subsection A2.
grow large and/or more accurate reporting practices for larger countries. The takeaway from these
two exercises is twofold: First, empirically, the approximation of net capital flows via net current
account flows seems to be good (consider that the exercise already excludes cases where they are
equal by construction), although not perfect. This lends credibility to ’indirectly’ estimating capital
account flows by equating the balances via the addition of a balancing item. Second, deviations from
perfect equality are, in some cases, substantial but non-systematic and decreasing in gross flows.
In the construction of the data, I now copy the practice of equating current and capital accounts
via the computation of an additional balancing item.34 To do so, I fill the capital account balance
with the negative current account balance, and compute the additional gross flow item that fulfills
the condition Credit − Debit = Capital Balance = Current Balance ∗ −1. Out of a total of 363
country-year observations in the final regression sample, 74 include a balancing item computed by
me. In addition to this, 71 observations contain equal balances in the raw data, giving a total of 145
observations in which the balances are made to fit by construction. The country-year observations
for which this applies are listed in Table A2.1, where ”Equal Balance” refers to all observations with
equal balances and ”Set Equal Balance” refers to the subset where this equality is set by me.
The second step involves computing additional capital account entries as the residual of available
information. This might involve the computation of a particular type of flow, like e.g. in the case
of the Netherlands which reports only total flows in the final table, but lists long-term flows as an
intermediate item earlier on, the difference of which I record as short-term flows in my data. It might
34 For reference: the largest country that follows this practice in the original data is Germany. However, after the NSDAP’s
rise to power, Germany stops reporting to the League after 1935 and when the League of Nations tries to infer the statistics
itself, it ceases to employ balancing items.
A3
Table A2.1: Coverage of BoP capital flow data and data construction details
Country
Regression Sample
Equal Balances
Set Equal Balances
Computed Residuals
Set Equal Balances
and Computed
Residuals
Albania
Argentine
Australia
Austria
Belgium & Lux.
Bulgaria
Canada
China
Czechoslovakia
Denmark
Dutch East Indies
Estonia
Finland
France
Germany
Greece
1927-1933
1922-1938
1923-1936
1926-1936
1930-1937
1925-1936
1926-1938
1929-1938
1926-1937
1924-1938
1926-1938
1926-1937
1927-1938
1922-1937
1925-1937
1930-1938
1928, 1929
1931
1923-1927
1926-1936
1930-1933, 1935-1937
1925-1936
1928, 1929
1929
1933-1937
1923-1927
1926-1929, 1932-1936
1935-1937
1929
Hungary
India
Iraq
Ireland
Japan
Latvia
Lithuania
Netherlands
New Zealand
Norway
Poland
Romania
Sweden
Thailand
Turkey
Union of South Africa
United Kingdom
United States
Yugoslavia
1924-1936
1924-1938
1931-1934
1925-1938
1926-1936
1924-1937
1925-1937
1924-1937
1924-1937
1924-1938
1924-1937
1927-1930
1924-1937
1933-1937
1927-1933
1924-1937
1925-1937
1923-1937
1927-1935
1923-1927
1935-1937
1923-1927
1935-1937
1926-1935
1929, 1930, 1933-1937 1929, 1930, 1933-1937 1929, 1930, 1933-1937
1929-1938
1928
1928
1931, 1932
1922-1931, 1933-1937 1922-1931, 1933-1937 1922-1931, 1933-1937 1922-1931, 1933-1937
1925-1937
1936, 1937
1936, 1937
1936, 1937
1930-1932, 1935, 1937,
1938
1925-1930, 1932
1925-1930, 1932
1925-1938
1925-1930, 1932
1924-1929
1924-1926, 1928, 1929
1935
1925-1930, 1936
1930
1926-1929
1926-1929, 1934-1937
1926-1929
1925-1930, 1936
1930
1925-1930, 1936
1925-1930, 1936
1930-1933
1924-1932
1925-1937
1930-1933
1925-1937
1925-1937
1925-1937
1927-1935
1927-1929, 1935
1927-1929, 1935
1927-1929, 1935
Notes: This table shows the sample coverage of capital flows, together with details on their construction. Column (3) lists all observations
for which the capital account balance is equal to the inverse current account balance. Column (4) shows the subset of observations where I
have set the two to be equal. Column (5) lists all instances that involve the computation of an additional capital flow item not found in the
original publications. Column (6) lists observations for which I have set the balances to be equal and computed residual positions.
also involve computing the described gross balancing item which is added to either gross inflows or
gross outflows to equate the balances (in some circumstances, when either only inflows or outflows
are known, it might also result in the creation of the missing item, equivalent to adding to a position
that was previously 0). As long-term flows tend to be more reliable and complete in the data, I
add the computed item to short-term flows when breaking down flows by their maturity.35 The
column headed ”Computed Residuals” lists the 106 observations for which any additional figure has
been computed as a residual of the known items and that is not found in the original publications.
The rightmost column lists the 68 observations for which I have set the balances to be equal and
computed residual positions, signifying the most extensive form of data alteration.
Finally, as not only the capital account but also the yearly coverage of countries is potentially
35 The precise distinction would thus be: flows which I know to be long-term vs. short-term + flows of unknown maturity.
A4
Figure A2.4: Data validation, comparison to Jordà et al. (2011)
-3
-1.5
0
1.5
3
BoP Current Account Balance
-3
-1.5
0
1.5
BoP Capital Account Balance
3
2
-2
-1
0
1
JST Current Account Balance
2
1
0
-2
FIN1928
-3
-3
FIN1928
-1
0
FIN1933
-1.5
0
-1.5
CAN1925 GBR1923
DEU1931 SWE1934
FRA1930
SWE1929
DNK1938
DNK1937 FIN1934
JPN1935
JPN1936
AUS1932
SWE1927
NLD1928
NLD1938
CAN1936
AUS1934
NLD1927
FRA1929
CAN1926
USA1921
NOR1936
GBR1928
NOR1933
NOR1938
FIN1932DEU1926
AUS1933
FRA1925
USA1938
CAN1924
DNK1935
NOR1937 FRA1924
DEU1930
FIN1931
NLD1937
FRA1927
FRA1926 FRA1928
NOR1934
NLD1926
NOR1932
GBR1927
FIN1935
SWE1937
GBR1929
CAN1937
FIN1936
SWE1933 JPN1929
AUS1925
AUS1936
GBR1924CAN1935
NLD1925
NOR1935
FRA1923
NLD1936
DNK1929
USA1924
AUS1931
DEU1936
JPN1930
JPN1933
JPN1932
DEU1932
CAN1923
AUS1935
DEU1937
USA1928
DEU1933
DNK1925
DNK1932
DNK1928
NOR1926
CAN1927
DNK1926
CAN1934
DNK1927
USA1927
JPN1934
GBR1925
SWE1936
USA1929
CAN1938
SWE1926
DNK1936
USA1930
GBR1930
CAN1928
NLD1929
DNK1933
USA1925
DEU1935
SWE1930 NLD1930
CAN1922
DEU1929
CAN1933
USA1922
GBR1935
FRA1922
FIN1924
SWE1932
NOR1929
FIN1930
FRA1931
NLD1924
NOR1927
NLD1935
USA1931
FIN1925
USA1926
USA1923
SWE1925
AUS1923
FIN1937
SWE1938
JPN1931 SWE1928
FIN1926
NOR1930
DEU1934
NOR1925
FIN1927
GBR1933
AUS1926
DNK1934
DNK1924
JPN1928
USA1934
JPN1927
FIN1923
FRA1935
CAN1932
GBR1934 GBR1936
DNK1930
GBR1926
FIN1938
AUS1924
FRA1934
USA1937
AUS1929
AUS1928
USA1933 NOR1931
NOR1928
CAN1921DEU1928
GBR1937
FRA1938
NLD1931
GBR1932
FRA1932
GBR1938
SWE1935
CAN1931
SWE1924 USA1936
FRA1936
FIN1929
NLD1934
USA1935
DNK1931
USA1932
NLD1932
AUS1927
FRA1933
DEU1925
GBR1931NOR1924
FRA1937
CAN1929
SWE1931
JPN1926CAN1930
DEU1927
AUS1930
NLD1933
1.5
FRA1930
USA1938
NLD1937
NLD1926
Capital Account
JST Current Account Balance
JST Current Account Balance
CAN1925
GBR1923
SWE1934
DEU1931
SWE1929
DNK1938
DNK1937
FIN1934
JPN1935
JPN1936
AUS1932
FIN1933
NLD1928AUS1934 CAN1936
SWE1927
NLD1938
FRA1929 NLD1927
CAN1926
USA1921
NOR1936
GBR1928
NOR1938
NOR1933
FIN1932
AUS1933
FRA1925
CAN1924
DEU1926
DNK1935
NOR1937
DEU1930
FRA1924
FIN1931
FRA1927
FRA1928FRA1926
NOR1934
NOR1932
GBR1927
FIN1935
SWE1937
GBR1929
CAN1937
FIN1936
SWE1933
AUS1925 JPN1929
CAN1935 AUS1936
GBR1924
NLD1925
NOR1935
FRA1923
NLD1936
DNK1929 AUS1931
USA1924
DEU1936
JPN1933
JPN1932
DEU1932
CAN1923
AUS1935
DEU1937JPN1930
USA1928
DEU1933
DNK1925
DNK1928
DNK1932
NOR1926
CAN1927
DNK1926 DNK1927
CAN1934
USA1927
JPN1934
GBR1925
SWE1936
USA1929
CAN1938
SWE1926
DNK1936
USA1930
GBR1930
CAN1928
NLD1929
DNK1933
NLD1930
USA1925
DEU1935
SWE1930
CAN1922
DEU1929
CAN1933
USA1922
GBR1935
FRA1922
FIN1924
SWE1932
NOR1929
FIN1930
FRA1931
NLD1924
NOR1927
NLD1935
USA1931
FIN1925
USA1926
USA1923
SWE1925
SWE1928
AUS1923
FIN1937
SWE1938 NOR1930
JPN1931
FIN1926
DEU1934
NOR1925
FIN1927
GBR1933
AUS1926
DNK1934
DNK1924
JPN1928
USA1934
JPN1927
FIN1923
CAN1932
DNK1930
GBR1926
FIN1938
AUS1924 FRA1935
GBR1936 GBR1934
FRA1934
USA1937
AUS1929
AUS1928
NOR1931 USA1933
NOR1928
DEU1928
CAN1921
GBR1937
FRA1938
NLD1931
GBR1932
FRA1932
GBR1938
SWE1935
USA1936
CAN1931
SWE1924
FIN1929
NLD1934FRA1936
USA1935
DNK1931NLD1932
USA1932
AUS1927 FRA1933
NOR1924DEU1925 GBR1931
FRA1937
SWE1931 CAN1929
JPN1926
DEU1927
CAN1930
AUS1930
NLD1933
1.5
JST Current Account Balance
Current Account
3
Capital Account
3
Current Account
-2
-1
0
1
2
BoP Current Account Balance
-2
-1
0
1
2
BoP Capital Account Balance
Notes: This figure compares the newly collected capital flow data from the BoP to net capital flows (the current account) collected by Jordà
et al. (2011). The left two panels plot the current account balance from Jordà et al. (2011) against the current and capital account balance
from the BoP, respectively, displaying individual country-year observations. The solid line shows the linear relationship, while the dotted
line represents a non-parametric estimate. The right two panels plot the same relationship in binned scatterplots absorbing country fixed
effects.
incomplete, I interpolate missing years in the data using simple linear interpolation. Due to the
publications being at a yearly frequency, and each publication usually containing several years of
data, the absence of one country from a specific publication does not usually lead to a gap in the
data, and interpolation is not frequent. Only 12 out of 363 observations in the final regression sample
include an interpolated value.36
How well does this newly constructed dataset correspond to previously available data? To answer
this question, I now compare my data to the net capital flows collected by Jordà et al. (2011) for
a subset of my sample. The left two panels of Figure A2.4 scatter the current and capital account
balances from my sample against the current account balance collected by Jordà et al. (2011). The
solid line is based on a linear regression, and the dashed line on a non-parametric estimate of the
relationship. Generally, the link is as expected, being positive for the current and negative for the
capital account balance. The relationship for the current account seems to be closer, though not
identical. For the capital account, it seems to be especially low values in the capital account balance
(likely caused by under reported inflows) that deviate from the estimates in Jordà et al. (2011),
indicated by the downward deviation of the non-parametric from the linear relationship. This picture
is confirmed in the right two panels using binned scatterplots and absorbing country fixed effects.
A2-1
A Note on interpretation
While the BoP provides relatively detailed information on the movement of capital, it reports all
cross-border movements of financial assets indiscriminately of their origin and does not directly map
onto modern definitions of gross capital flows. The standard definition of a gross capital inflow is
the net change of gross domestic assets held by foreigners, while a gross capital outflow is the net
change of gross foreign assets held domestically. The BoP deviates from this definition as it contains
any transaction across borders and thus includes partially offsetting values. For example, the sale of
36 These are: Austria (1930-1931), China (1931-1932), Hungary (1925), Netherlands (1925), Yugoslavia (1930-1934).
A5
a domestic or foreign asset by a domestic agent to a foreign agent is considered a gross inflow in the
BoP (exchange of asset ownership for capital inflow). In modern classifications, however, the sale of
a domestic asset to a foreigner is reported as an increase in gross foreign liabilities, while the sale
of a foreign asset is a decrease in gross foreign assets. From a practical perspective, a gross inflow
in the BoP thus represents a decrease in the net claims a country has on the rest of the world, as
foreign borrowing (i.e., selling domestic assets to foreigners) increases foreign claims on the domestic
country, while selling foreign assets to foreigners decreases claims the country has on the rest of the
world.
Generally, the BoP data in this paper is normalized using z-score normalization at the country
level. This approach has the advantage that, unlike logs, it easily handles negative values in the
balance, reduces noise, and simplifies interpretability through unit standard deviation. It is also
uniformly available and does not require the collection of an additional scaling variable (which
might come with its own set of assumptions). However, this approach does not account for relative
differences in capital movements between countries, as the mean value around which the z-score
is centered may vary significantly, for instance, in relation to GDP. To address this, I also conduct
robustness exercises in which capital flows are scaled with nominal GDP. As backward estimations of
nominal GDP are notoriously difficult and introduce additional assumptions, I also employ a measure
from the same source as the BoP to assess the relative size of flows with information contemporaries
would have had: domestic bank balance sheet size.
A final note on the interpretation of the constructed data: While the precise composition of gross
capital flows should be taken with a few grains of salt, my analysis does not rely on it. My baseline
relies only on total gross inflows and outflows, which, despite including some approximation,
show a high degree of internal consistency and seem to correspond, on the whole, to what is
known from earlier data. The more dubious classification into long-term and short-term flows is
consequently mainly used in the description of capital flows and robustness exercises, rather than as
an independent variable in the baseline regressions. Ultimately, I do not believe that previous authors
have overstated the shortcomings of the BoP data, but rather that they might have underestimated
what we can take away from it despite these shortcomings. After all, if the data were systemically
misreported, the open question would be why gross capital flows perform so much better than net
flows in predicting economic outcomes. And if we had perfect data, would it perform better still?
A6
A2-2
Complementary Data
To link the BoP to the business cycle, it is complemented with Maddison-style GDP estimates from
Bolt and van Zanden (2020), GDP estimates for the Baltic states collected by Klimantas and Zirgulis
(2020) and Norkus and Markevičiūtė (2021), GDP growth rates from Baron et al. (2021), and economic
activity indicators (EAI) constructed by Albers (2018). Growth variables are harmonized across
sources by being expressed in log-differences. Results are generally reported for the full combined
sample, but robustness checks also include a breakdown by source of GDP data, to confirm that
results are not driven by a single source affecting the aggregate results.
The baseline financial crisis indicator is the crisis chronology of Baron et al. (2021), supplemented
by the indicator for banking crises by Reinhart and Rogoff (2009) and Grossman (1994) when
countries are not covered. While the crisis episodes compiled by these authors partially diverge in
their definitions of crises, they together provide comprehensive coverage of my sample. For full
transparency, an overview of the crises from different sources and the final coverage is given in
Table A2.3. Results are also estimated for the individual subsets, ensuring that no single chronology
is driving the aggregate results. For overlapping countries, this also closely corresponds to the
chronology of Da Rocha and Solomou (2015), with my classification generally being slightly more
liberal than theirs due to the inclusion of several types of crises from the BVX chronology and using
banking crises from Reinhart and Rogoff while they explicitly focus on systemic crises.
Gold Standard indicators are similarly compiled from Reinhart and Rogoff (2009) and supplemented by Eichengreen (1992) and Wandschneider (2008). The capital account openness measure is
based on Quinn (2003). For the construction of the instrumental variable, bilateral data for portfolio
investments of the United States are collected from Dickens (1931, 1930) and Lewis and Schlotterbeck
(1938). Bank balance sheet data, used for scaling and in the chapter on maturity mismatches, is
collected from League of Nations (1931-1940). The data for gold and foreign asset reserves, used in
the chapter on the Gold Standard, is compiled from League of Nations (1936-1939b) and League of
Nations (1921-1939c). Summary statistics for the main interwar variables are shown in Table A2.2
A7
Table A2.2: Summary statistics of main interwar variables
Mean
Median
Std. Dev.
Min
Max
Obs
Panels
GDP 1 Year log Growth
0.024
0.027
0.052
-0.168
0.212
475
35
GDP 2 Year log Growth
0.048
0.052
0.082
-0.242
0.322
457
35
GDP 3 Year log Growth
0.071
0.080
0.108
-0.329
0.444
430
35
GDP 4 Year log Growth
0.090
0.097
0.127
-0.364
0.567
400
35
GDP 5 Year log Growth
0.106
0.099
0.142
-0.351
0.690
369
35
Capital, Balance (USD)
0.003
0.002
0.191
-0.951
1.036
475
35
Capital, Debit (USD)
0.173
0.034
0.410
0.000
3.656
451
35
Capital, Credit (USD)
0.179
0.050
0.389
0.000
2.602
451
35
Capital, Balance t to t-2 (USD)
-0.002
0.003
0.516
-3.829
2.483
406
35
Capital, Debit t to t-2 (USD)
0.541
0.114
1.238
0.000
9.477
381
35
Capital, Credit t to t-2 (USD)
0.560
0.154
1.144
0.000
7.439
381
35
Capital, Balance (% GDP)
0.293
0.068
1.401
-6.281
9.531
433
32
Capital, Debit (% GDP)
1.278
0.811
1.373
0.000
6.417
409
32
Capital, Credit (% GDP)
1.607
0.949
1.683
0.000
10.440
409
32
Combined Interwar Crisis
0.088
0.000
0.284
0.000
1.000
475
35
Gold Standard
0.411
0.000
0.492
0.000
1.000
475
35
Gross Exposure Dummy (GED)
0.229
0.000
0.421
0.000
1.000
475
35
Gross Foreign Assets Dummy (GFA)
0.251
0.000
0.434
0.000
1.000
475
35
Crisis × GED
0.025
0.000
0.157
0.000
1.000
475
35
Crisis × GFA
0.036
0.000
0.186
0.000
1.000
475
35
Capital Account Openness
70.593
62.500
28.843
12.500
100.000
295
21
Gold Ratio
0.430
0.395
0.238
0.006
1.206
423
31
Foreign Asset Ratio
0.227
0.162
0.221
0.002
1.537
414
31
Bank Balance Sheet Size (USD)
3.523
0.757
10.151
0.014
63.734
385
29
Notes: This table shows summary statistics of the main variables. Output growth variables are compiled from various sources and
jointly expressed in cumulative log differences over the indicated period. Variables from the Balance of Payments are expressed in
billion USD and relative to GDP. Crises and Gold Standard variables are included as dummy variables, with their mean expressing the
unconditional sample frequency of crises or Gold Standard adherence. GED and GFA refer to the constructed dummy variables, capturing
the accumulation of foreign credit or foreign assets. Their interaction refers to the the respective subset of crisis. The capital account
openness measure is based on Quinn (2003) and reports the unconditional capital account openness (for reference: during the Gold
Standard, the average openness score is 93 versus only 49 outside of it). Gold and foreign asset ratios are relative to notes in circulation.
Bank balances sheet size is in billion USD. Reported statistics are the variable mean, median, standard deviation, minimum, maximum,
the total number of observations and the number of panels (countries) for which the variable is available.
A8
Table A2.3: Financial crises in sample
Country
Argentine
Australia
Austria
Belgium & Lux.
Bulgaria
Canada
China
Czechoslovakia
Denmark
Dutch East Indies
Estonia
Finland
France
Germany
Greece
Hungary
India
Ireland
Japan
Latvia
Lithuania
Netherlands
New Zealand
Norway
Poland
Romania
Sweden
Turkey
Union of South Africa
United Kingdom
United States
Yugoslavia
Baron, Verner, Xiong
Reinhart, Rogoff
Combined Interwar
1930, 1934
1931
1929, 1931
1931
1931, 1934
1931, 1932
1929, 1931
1931, 1934
-
1931, 1934, 1937
1930, 1934
1931
1929, 1931
1931
1931, 1934, 1937
1931
1931
1930, 1931
1931
1930, 1937
1925, 1929, 1930
1931
1931
1929
1927
1931
1927, 1931, 1936
1931, 1932, 1934
1931
1931
1930, 1931
1929, 1930
1931, 1932
1931
1931
1931
1930, 1937
1925, 1929, 1930
1931
1931
1929
1927
1931
1931, 1939
1930, 1931, 1932
1929, 1930, 1931, 1932
1931, 1932
1931, 1932
1929, 1930, 1931
1927
1931
1927, 1931, 1936
1931
1930, 1931
1929, 1930
1927, 1936
1931, 1932, 1934
1931
1931, 1932
1931
1929, 1930, 1931, 1932, 1933
Notes: This table shows the crisis dating from the chronologies of Baron et al. (2021) and Reinhart and Rogoff (2009). Crises are defined as
banking crises in Reinhart and Rogoff (2009), but, due to a more detailed breakdown, as either banking crises, banking panics, or narrative
banking crises in Baron et al. (2021). Despite Canada experiencing large losses in bank equity, no banking failures followed, consequently
excluding Canada from conventional crisis-dating (Bordo et al., 2001; Baron et al., 2021; Reinhart and Rogoff, 2009). The third column joins
the two crisis chronologies for a more comprehensive country coverage (RR is only employed when BVX is unavailable) and additionally
adds crisis dates from Grossman (1994). A blank space refers to the respective country not being covered by the chronology and a ”-”
indicates that the country is covered but experienced no crisis.
A9
A3.
Trends
Figure A3.5: Gross capital flows: comparison to net and gold flows
Gold and Current Account
.06
.1
.04
.25
.4
.08
.1
.55
Net / Gross Capital Flows
Gold / Gross Current Flows
.7
.12
Net and Gross Capital
12
14
16
18
Log Gross Capital Flows
20
22
12
14
16
18
20
Log Gross Current Flows
22
Notes: This figure relates gold and net flows to gross capital flows using binned scatterplots with 20 equally sized bins. The left panel
shows the ratio of absolute net flows to gross capital flows against total log gross capital flows. The right panel does so for the ratio of
gross gold to gross current account flows against log gross capital flows.
Capital Account
25
Current Account
20
Merchandise
Gold
0
5
20
0
1924
Total Capital
Long-Term Flows
Short-Term Flows
Billion USD
10
15
Total Current
Interest and Dividends
Services / Other
Billion USD
40
60
80
100
Figure A3.6: Trends in gross Balance of Payments flows
1927
1930
1933
1936
Year
1924
1927
1930
1933
1936
Year
Notes: This figure shows in the left panel the total gross flows (Credit + Debit) for the individual parts of the current account in billion
USD. Flows in trade (purple) make up by far the largest part, with secondary incomes (green), services (gray) and gold (gold) making up
the remainder. The right panel shows this decomposition for the capital account. While long-term capital flows (red) generally make up
the largest share, short-term flows (gray) make up a sizable portion and gain in importance around the Great Depression.
A10
A4.
Capital flows and business cycle dynamics, robustness
Tables A4.4, A4.5, A4.6, and Figure A4.7 test the robustness of the relationship between capital inflows
and adverse outcomes by using 5 year sums of capital inflows, estimating responses separately for
different GDP sources, comparing it to other prominent explanatory variables for the business
cycles, and running country-level regressions, respectively. For a more detailed description see
subsection 4.3.
Table A4.7 checks the baseline specification for robustness against potentially biasing factors. I
first address the concern that the long downturn of the Great Depression might be the sole driver of
the observed relationship and split the sample in 1930 (meaning the last capital flows included are in
1928, so that the two year forecast stops in 1930). The relationship holds in both sub-samples, but is
notably weaker in the smaller earlier sample. Is only a small group of core countries producing the
results? To answer this question I estimate coefficients for the core countries of North America and
Europe37 and all other countries separately. Again, the results hold in both groups. To account for
present time net availability of capital, I split the sample along the current account being positive
or negative in year t, which produces virtually identical coefficients. Lastly, the link between gross
capital inflows and output might be non-linear, with one tail of the distribution accounting for all
variation in outcomes. I interact credit with a dummy for credit between t and t − 2 being above or
below zero38 and show that the relationship is, in fact, close to linear.
To account for the possibility of a common global business cycle, Table A4.8 adds year fixed effects
to my baseline specification (columns (1), (4), and (7) show baseline coefficients for comparison). Note,
however, that this may underestimate the link between foreign credit and domestic business cycles if
country-specific inflows are driven by a common global capital supply component that is responsible
for part of the negative association with macroeconomic outcomes (Rey, 2013; Miranda-Agrippino
and Rey, 2020). This argument is discussed in section 7. While this reduces coefficient size, gross
foreign inflows remain significantly associated with lower future GDP growth. Ideally, I could fully
account for country-time specific factors using country×year fixed effects. However, as the data is
at the country-year level, this is not possible. As an alternative specification, columns (3), (6), and
(9) add country×period fixed effects to the model.39 Interestingly, including country×period fixed
effects in the model increases coefficients almost back to their previous level.
The normalized BoP variables do not consider differences in the relative size of capital flows
across countries, e.g. in relation to GDP. Table A4.9 replicates the baseline specification for variables
scaled by Maddison-style real GDP estimates in columns (1)-(4) and by domestic bank balance sheet
37 The United States, United Kingdom, France, Germany, Canada and the Netherlands
38 Since all variables are normalized, this is equivalent to credit growth being above or below mean growth.
39 Periods are defined following the interwar business cycle, with the first period going up until 1925 (roughly WWI
recovery), the second from 1926 to 1929 (expansion), the third from 1930 to 1933 (depression), and the fourth for years after
1934 (recovery from the Great Depression).
A11
size, collected from from League of Nations (1931-1940) in column (5)-(8). Across both specifications,
gross inflows retain their statistical significance and remain the most robust predictors of economic
slowdowns in the medium term.
Table A4.10 extends the analysis to financial and non-financial equity returns, as well as industrial
production. Equity returns are expressed as the cumulative percent change and industrial production
as a log change. Again, gross inflows robustly predict lower future returns and growth, respectively.
Interestingly, the effects are larger for non-financial equity relative to financial equity, and for
industrial production relative to aggregate GDP, corresponding to the results in Da Rocha and
Solomou (2015), and confirming the importance of declining industrial production during the Great
Depression.
Table A4.4: Capital flows and business cycle dynamics, 5 year cumulative capital flows
∆2 Yi,t+2
Σ4j=0 Balancei,t− j
(1)
(2)
-0.02∗∗∗
(0.01)
0.00
(0.01)
Σ4j=0 Crediti,t− j
-0.04∗∗∗
(0.01)
Σ4j=0 Debiti,t− j
R2
Country fixed effects
Lagged Growth
p-value, β Credit = β Balance
p-value, β Credit = β Debit
Observations
∆3 Yi,t+3
(3)
(4)
(5)
-0.03∗∗∗
(0.01)
-0.00
(0.01)
-0.04∗∗∗
(0.01)
-0.05∗∗∗
(0.01)
0.00
(0.01)
0.208
✓
✓
0.282
✓
✓
0.02
291
291
∆4 Yi,t+4
(6)
(7)
(8)
-0.03∗∗∗
(0.01)
-0.01
(0.01)
-0.05∗∗∗
(0.01)
-0.03∗∗
(0.01)
0.00
(0.01)
0.282
✓
✓
0.369
✓
✓
0.433
✓
✓
0.04
0.00
291
263
263
(9)
-0.04∗∗∗
(0.01)
0.00
(0.01)
0.433
✓
✓
0.616
✓
✓
0.636
✓
✓
0.21
0.00
263
234
234
0.635
✓
✓
0.01
234
Notes: This table regresses log GDP growth over varying time horizons on cumulative capital account flows summed from t to t − 4.
All specifications control for country fixed effects. Adjusting for longer time spans, lagged growth indicates three, four and five year
distributed lags of GDP growth, depending on the length of the forecast horizon. The reported p-value refers to a test for the equality of
coefficients. Standard errors in parentheses are dually clustered on country and year and *,**,*** indicates significance at the 0.1, 0.05, 0.01
levels, respectively.
A12
Table A4.5: Capital flows and business cycle dynamics, by source of GDP data
∆2 Maddisoni,t+2
Σ2j=0 Balancei,t− j
(1)
(2)
-0.02∗∗∗
(0.01)
0.01
(0.01)
Σ2j=0 Crediti,t− j
∆2 EAIi,t+2
(3)
-0.04∗∗∗
(0.01)
(5)
-0.00
(0.01)
-0.04∗∗∗
(0.01)
Σ2j=0 Debiti,t− j
R2
Country fixed effects
Lagged Growth
p-value, β Credit = β Balance
p-value, β Credit = β Debit
Observations
(4)
-0.03∗∗∗
(0.01)
(6)
-0.05∗∗
(0.02)
0.247
✓
✓
0.00
327
327
(7)
(8)
-0.02∗∗∗
(0.01)
0.01
(0.01)
-0.05∗∗∗
(0.02)
-0.00
(0.01)
0.126
✓
✓
∆2 Yi,t+2
-0.04∗∗∗
(0.01)
0.00
(0.01)
0.243
✓
✓
0.318
✓
✓
0.423
✓
✓
0.08
0.00
327
162
162
(9)
-0.04∗∗∗
(0.01)
0.00
(0.01)
0.424
✓
✓
0.123
✓
✓
0.233
✓
✓
0.01
0.00
162
363
363
0.230
✓
✓
0.00
363
Notes: This table presents estimation results from Equation 3 for different sources of GDP growth separately. The two largest contributors
to the total sample are Maddison style GDP estimates from Bolt and van Zanden (2020) and economic activity indicators from Albers
(2018). The combined sample also includes growth rates from Baron et al. (2021) and estimates for the Baltic states from Klimantas and
Zirgulis (2020); Norkus and Markevičiūtė (2021). The respective dependent variable is expressed as log growth from t to t + 2. The
independent variables are cumulative capital account flows in years t − 2 to t. All specifications control for country fixed effects and
lagged growth indicates a two year distributed lag of GDP growth. The reported p-value refers to a test for the equality of coefficients.
Standard errors in parentheses are dually clustered on country and year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels,
respectively.
Table A4.6: Capital flows and business cycle dynamics, comparison to other explanatory variables
∆2 Yi,t+2
(1)
Σ2j=0 Crediti,t− j
(2)
(3)
(4)
-0.04∗∗∗
(0.01)
Σ2j=0 Balancei,t− j
(5)
(6)
(0.01)
0.00
(0.01)
-0.02∗∗
(0.01)
Σ2j=0 Crediti,t− j × ∆2 Loansi,t−1
-0.01∗∗
(0.01)
-0.01∗
(0.01)
-0.02∗∗
(0.01)
-0.01∗∗
(0.01)
-0.07∗∗∗
(0.02)
Gold Standardi,t
-0.06∗∗
(0.03)
Crisisi,t
Gold Standardi,t × Crisisi,t
0.269
✓
✓
287
0.141
✓
✓
287
(8)
-0.03∗∗∗
(0.01)
-0.02∗∗∗
(0.01)
∆2 Domestic Loansi,t−1
R2
Country fixed effects
LDV
Observations
(7)
-0.04∗∗∗
0.119
✓
✓
287
0.317
✓
✓
287
0.198
✓
✓
287
0.106
✓
✓
287
-0.07∗∗∗
(0.02)
-0.03
(0.02)
-0.03∗∗∗
(0.01)
-0.01
(0.01)
-0.02
(0.03)
-0.03
(0.04)
0.219
✓
✓
287
0.346
✓
✓
287
Notes: This table presents estimation results from Equation 3. The dependent variable is log GDP growth over horizons t to t + h. The
independent variables are cumulative capital account flows from t − 2 to t and additional variables that have been used to explain interwar
business cycle dynamics, including domestic credit growth, the Gold Standard, and its interaction with a financial crisis indicator. All
specifications control for country fixed effects and two lags of GDP growth. Standard errors in parentheses are dually clustered on country
and year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
A13
Figure A4.7: Capital flows and business cycle dynamics, country level regression coefficients
Notes: This figure plots regression coefficients and 90% confidence intervals, using Newey-West standard errors with a lag length of four,
from individual time series regressions of log GDP growth from t to t + 2 on accumulated credit and debit positions from t − 2 to t. The
shown coefficients are βC , for capital account credit, with the specification ∆2 yi,t+2 = αi + βC Σ2j=0 Crediti,t− j + β D Σ2j=0 Debiti,t− j + ui,t+2
estimated on individual country sub-samples.
A14
Table A4.7: Capital flows and business cycle dynamics, sample splits, state dependence and linearity
∆2 Yi,t+2
Time Split
Σ2j=0 Crediti,t− j
Σ2j=0 Balancei,t− j
Country Split
State Dependence
Linearity
(1)
(2)
(3)
(4)
(5)
(6)
-0.02∗
-0.05∗∗∗
-0.07∗∗∗
-0.03∗∗∗
-0.04∗∗∗
-0.03∗∗∗
(0.01)
(0.01)
(0.02)
(0.01)
(0.01)
(0.01)
0.01
(0.01)
0.03∗
-0.00
(0.00)
0.01
(0.01)
0.00
(0.01)
-0.01
(0.01)
(0.02)
(7)
0.01
(0.01)
Σ2j=0 Crediti,t− j × 1(> 0)
-0.04∗
(0.02)
Σ2j=0 Crediti,t− j × 1(< 0)
-0.04∗∗∗
(0.01)
R2
Country fixed effects
Lagged Growth
Sample
Countries
Current Account
Observations
0.270
✓
✓
pre 1930
0.338
✓
✓
post 1930
89
0.442
✓
✓
0.188
✓
✓
Core
Non-Core
73
290
269
0.170
✓
✓
0.161
✓
✓
0.233
✓
✓
Positive
173
Negative
187
363
Notes: This table presents estimation results from Equation 3, including sample splits and checks for state dependencies. The dependent
variable is log GDP growth from t to t + 2. The independent variables are cumulative capital account flows in years t − 2 to t. Core
countries are defined as the largest economies in North America and Europe: the United States, the United Kingdom, France, Germany,
Canada and the Netherlands. All specifications control for country fixed effects and lagged growth indicates a two year distributed lag of
GDP growth. Standard errors in parentheses are dually clustered on country and year and *,**,*** indicates significance at the 0.1, 0.05,
0.01 levels, respectively.
Table A4.8: Year and country×period fixed effects
∆2 Yi,t+2
∆3 Yi,t+3
∆4 Yi,t+4
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
-0.04∗∗∗
-0.01∗∗
-0.03∗∗∗
-0.06∗∗∗
-0.02∗∗
-0.04∗∗∗
-0.06∗∗∗
-0.02∗∗
(0.01)
(0.01)
(0.01)
(0.01)
(0.01)
(0.01)
(0.01)
(0.01)
-0.04∗∗∗
(0.01)
Σ2j=0 Debiti,t− j
0.00
(0.01)
0.01
(0.01)
0.01∗∗
(0.01)
0.00
(0.01)
0.01
(0.01)
0.02∗∗
(0.01)
-0.00
(0.01)
0.01
(0.01)
0.01
(0.01)
R2
Country fixed effects
LDV
Year fixed effects
Country × Period fixed effects
Observations
0.200
✓
✓
✓
0.467
✓
✓
✓
0.298
✓
✓
✓
0.558
✓
✓
✓
0.651
✓
✓
✓
363
336
336
0.702
✓
✓
✓
✓
336
0.426
✓
✓
✓
363
0.593
✓
✓
✓
✓
363
306
306
0.737
✓
✓
✓
✓
306
Σ2j=0 Crediti,t− j
Notes: This table presents estimation results from Equation 3, including year fixed effects in columns (2), (5), and (8), and country×period
fixed effects in columns (3), (6), and (9). Columns (1), (4), and (7) show the baseline coefficients for comparison. The dependent variable is
log GDP growth from t to t + 2. The independent variables are cumulative gross capital flows in years t − 2 to t. Periods are defined
following the interwar business cycle, with the first period going up until 1925 (roughly WWI recovery), the second from 1926 to 1929
(expansion), the third from 1930 to 1933 (depression), and the fourth for years after 1934 (recovery from the Great Depression). All
specifications additionally control for country fixed effects, and a two-year distributed lag of GDP growth indicated by LDV. Standard
errors, shown in parentheses, are dually clustered on country and year, and *, **, *** indicate significance at the 0.1, 0.05, and 0.01 levels,
respectively. See Appendix A4 for a discussion of these specifications.
A15
Table A4.9: Capital flows and business cycle dynamics, scaling capital flows by GDP and BBS
∆2 Yi,t+2
Scaled by GDP
(1)
Σ2j=0 Balancei,t− j
(2)
Scaled by BBS
(3)
-0.74∗∗∗
(0.22)
Σ2j=0 Crediti,t− j
-0.99∗∗∗
(0.32)
Σ2j=0 Debiti,t− j
R2
Country fixed effects
Lagged Growth
Observations
-0.99∗∗∗
(0.31)
(4)
(5)
(6)
-0.06
(0.30)
-0.03∗∗∗
(7)
-0.01
(0.01)
(0.01)
-0.94∗
(0.49)
-0.06∗∗∗
(0.02)
0.02
(0.31)
0.166
✓
✓
330
0.197
✓
✓
330
(8)
-0.06∗∗∗
(0.02)
-0.05∗∗
(0.02)
0.01
(0.01)
0.197
✓
✓
330
0.197
✓
✓
330
0.155
✓
✓
306
0.173
✓
✓
306
0.174
✓
✓
306
0.174
✓
✓
306
Notes: This table presents estimation results from Equation 3. The dependent variable is log GDP growth over horizons t to t + h. The
independent variables are cumulative capital account flows from t − 2 to t. Capital flow variables have been scaled by nominal GDP from
Bolt and van Zanden (2020) in columns (1)-(4) and by domestic bank balance sheet size from League of Nations (1931-1940) in columns
(5)-(8) All specifications control for country fixed effects and two lags of GDP growth. Standard errors in parentheses are dually clustered
on country and year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
Table A4.10: Capital flows, equity returns and industrial production
∆2 FI − Equityi,t+2
Σ2j=0 Balancei,t− j
(1)
(2)
-0.01
(0.02)
0.05
(0.03)
Σ2j=0 Crediti,t− j
-0.09∗∗
(0.04)
Σ2j=0 Debiti,t− j
R2
Country fixed effects
LDV
Lagged GDP Growth
Observations
∆2 NF − Equityi,t+2
(3)
(4)
(5)
-0.07∗∗
(0.03)
0.01
(0.04)
-0.06∗
(0.03)
-0.14∗∗
(0.07)
-0.04
(0.03)
0.053
✓
✓
✓
218
0.111
✓
✓
✓
218
0.109
✓
✓
✓
218
∆2 Ind. Prod.i,t+2
(6)
(7)
(8)
-0.06∗∗
(0.02)
-0.00
(0.03)
-0.13∗∗∗
(0.05)
-0.09∗∗
(0.04)
0.00
(0.03)
0.219
✓
✓
✓
227
0.283
✓
✓
✓
227
0.282
✓
✓
✓
227
(9)
-0.10∗∗∗
(0.03)
0.02
(0.02)
0.136
✓
✓
0.240
✓
✓
0.247
✓
✓
180
180
180
Notes: This table presents estimation results for alternative indicators of economic performance. Columns (1) to (6) show results for
cumulative equity returns for financial and non-financial corporations between t and t + 2 using the data from Baron et al. (2021), where
FI and NF refer to the equity returns of financial and non-financial institutions respectively. The independent variables are cumulative
capital flows from t − 2 to t. Columns (7) to (9) estimate the same regression for the log change in industrial production between t and
t + 2. All specifications control for country fixed effects and two distributed lags of the respective dependent variable as well as GDP
growth. *, **, *** indicate significance at the 0.1, 0.05, 0.01 levels, respectively.
A16
A5.
Capital flows and financial fragility
Figure A5.8: Capital flows and financial crises, conditional probability
1
2
3
4
5
Quintile, cumulative flows t-2 to t
12
0
4
8
12
8
4
0
0
4
8
12
Crisis frequency in year t (%)
16
Debit
16
Credit
16
Balance
1
2
3
4
5
Quintile, cumulative flows t-2 to t
1
2
3
4
5
Quintile, cumulative flows t-2 to t
Notes: This figure shows the relationship between cumulative capital flows between t − 2 and t and financial crisis frequencies for the year
t. Observations are sorted into five equal-sized quintiles according to the volume of cumulative capital flows between t − 2 and t. Vertical
bars indicate the frequency of financial crises in year t for each of these bins. The conditional frequency of financial crises can be seen to
peak in the highest quintile of gross foreign credit.
Table A5.11: Capital flows predicting financial crises, Reinhart and Rogoff (2009) crisis chronology
RR Crisisi,t+1
Σ2j=0 Balancei,t− j
(1)
(2)
0.02
(0.02)
-0.05∗
Σ2j=0 Crediti,t− j
(4)
(0.03)
0.10∗∗∗
(0.03)
Σ2j=0 Debiti,t− j
AUC
s.e.
Lagged Growth
Country fixed effects
Observations
(3)
0.06∗∗
(0.02)
(5)
(6)
0.04
(0.04)
-0.07
(0.05)
0.06∗∗∗
(0.02)
0.15∗∗∗
(0.06)
0.03
(0.02)
(7)
(8)
0.09∗∗∗
(0.03)
0.09∗∗∗
(0.03)
0.03
(0.03)
0.71
0.05
✓
0.77
0.04
✓
0.77
0.04
✓
0.76
0.04
✓
305
305
305
305
0.76
0.05
✓
✓
203
0.81
0.04
✓
✓
203
0.80
0.04
✓
✓
203
0.80
0.04
✓
✓
203
Notes: The table shows estimation results of a probit model for Reinhart and Rogoff (2009) financial crises, reporting mean marginal effects.
The independent variables are cumulative flows recorded in the capital account of the BoP in year −2 to t. AUC is the area under the
ROC-Curve, below it is its standard error. Standard errors in parentheses are clustered on country level and *,**,*** indicates significance
at the 0.1, 0.05, 0.01 levels, respectively.
A17
A6.
Increasing and decreasing recession severity:
Figure A6.9: Capital inflows and recession severity
Average Gross Inflows Around Crises
-.2
Poland
0
0
Finland
-.5
Union
of South Africa
Bulgaria
Australia
Denmark
Latvia
Estonia
Hungary
India
Norway
United
Kingdom
Ireland
Sweden Romania
New Zealand
Argentine
France
Netherlands
Dutch East Indies
Germany
Yugoslavia
Czechoslovakia
Lithuania
-.1
Mean Capital Inflows
.1
Albania
.5
Japan
Turkey
Austria
Recession > Median
Recession < Median
-1
United States
-.3
Log GDP Growth 1930 to 1933
.2
1
Gross Inflows and the Great Depression
Canada
-.5
0
.5
1
1.5
2
-3
-2
-1
0
1
2
3
Gross Foreign Credit 1927 to 1930
Crisis in t = 0
Notes: This figure relates gross capital inflows to recession severity. The left panel scatters normalized gross capital inflows between 1927
and 1930 against log GDP growth between 1930 and 1933, producing a visibly negative relationship. To account for heterogeneous crisis
starting dates, the right panel splits the sample along the median recession severity in the first 3 years after a crisis, using the combined
interwar crises chronology. It plots the average gross capital inflows in the 6-year window around crises for both groups, showing that
before a crisis, gross inflows are consistently higher in countries with larger than median recessions, but lower three years afterwards.
Table A6.12: GDP growth, crises and exposure to gross capital inflows, alternative GED2 measure
∆2 Yi,t+2
(1)
(3)
(4)
(5)
(6)
(7)
(8)
(0.02)
-0.00
(0.02)
0.03
(0.04)
0.02
(0.04)
-0.05∗∗
(0.02)
-0.01
(0.02)
0.02
(0.04)
0.02
(0.04)
-0.04∗∗
(0.01)
-0.03∗∗
(0.01)
-0.00
(0.01)
-0.01
(0.01)
-0.04∗∗
(0.02)
-0.04∗∗
(0.02)
-0.01
(0.01)
-0.01
(0.01)
-0.06∗∗
(0.02)
-0.07∗∗
(0.04)
-0.07∗
(0.04)
-0.05∗∗
(0.02)
-0.06∗
(0.04)
-0.06∗
(0.04)
Σ2j=0 Crediti,t− j
-0.04∗∗∗
(0.01)
-0.03∗∗
(0.01)
-0.04∗∗∗
(0.01)
-0.04∗∗∗
(0.01)
Σ2j=0 Balancei,t− j
0.00
(0.01)
0.00
(0.01)
0.00
(0.01)
0.00
(0.01)
Crisisi,t
GED2i,t
-0.05∗∗
(2)
Crisisi,t × GED2i,t
-0.03∗
(0.02)
Gold Standardi,t
R2
Country fixed effects
Lagged Growth
Crisis in Sample
Observations
0.122
✓
✓
0.126
✓
✓
0.254
✓
✓
0.276
✓
✓
342
342
342
342
-0.02
(0.02)
0.164
✓
✓
✓
241
0.168
✓
✓
✓
241
0.317
✓
✓
✓
241
0.321
✓
✓
✓
241
Notes: This table re-estimates Equation 6, with an alternative definition for foreign capital exposure prior to crisis. The alternative GED2dummy is defined as the first lag of the baseline credit variable ∑2j=0 Crediti,t− j being in the top 80%. The dependent variable is log GDP
growth over the period t to t + 2. The independent variables are financial crises in year t, the GED2-variable capturing exposure to capital
inflows, the baseline BoP-variables accumulated over t to t − 2 and a Gold Standard indicator. All specifications additionally control for
country fixed effects and a two year distributed lag of GDP growth. Standard errors in parentheses are dually clustered on country and
year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
A18
Table A6.13: GDP growth, crises and exposure to gross capital inflows, Reinhart and Rogoff (2009) crisis chronology
∆2 Yi,t+2
(1)
(2)
(3)
(4)
(5)
0.04∗∗∗
-0.04∗∗
(0.01)
0.01
(0.01)
0.04∗∗∗
(0.01)
(0.01)
-0.01
(0.02)
0.01
(0.01)
0.04∗∗∗
(0.01)
0.04∗∗∗
(0.01)
-0.13∗∗∗
(0.04)
-0.13∗∗∗
(0.03)
-0.12∗∗∗
(0.03)
Σ2j=0 Crediti,t− j
-0.05∗∗∗
(0.01)
Σ2j=0 Balancei,t− j
0.01
(0.01)
Crisisi,t
GEDi,t
-0.03∗∗
Crisisi,t × GEDi,t
(7)
(8)
(0.02)
0.01
(0.01)
0.03∗∗
(0.01)
0.03∗∗
(0.01)
-0.01
(0.02)
0.01
(0.02)
0.04∗∗
(0.02)
0.04∗∗
(0.02)
-0.13∗∗∗
(0.04)
-0.12∗∗∗
(0.04)
-0.12∗∗∗
(0.04)
-0.04∗∗∗
(0.01)
-0.05∗∗∗
(0.01)
-0.04∗∗∗
(0.01)
0.01
(0.01)
0.01
(0.01)
0.01
(0.01)
-0.03∗
(0.02)
Gold Standardi,t
R2
Country fixed effects
Lagged Growth
Crisis in Sample
Observations
(6)
0.085
✓
✓
0.147
✓
✓
0.315
✓
✓
0.338
✓
✓
276
276
276
276
-0.02
(0.02)
0.100
✓
✓
✓
212
0.177
✓
✓
✓
212
0.350
✓
✓
✓
212
0.359
✓
✓
✓
212
Notes: This table presents estimation results from Equation 6. The dependent variable is log GDP growth over the period t to t + 2. The
independent variables are financial crises in year t (using only Reinhart and Rogoff (2009) crisis dating), the GED-variable capturing
exposure to large capital inflows, the baseline BoP-variables accumulated over t to t − 2 and a Gold Standard dummy variable. All
specifications additionally control for country fixed effects and a two year distributed lag of GDP growth. Standard errors in parentheses
are dually clustered on country and year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
Table A6.14: Crisis moderation via foreign asset accumulation
∆2 Yi,t+2
Crises3
GFAi,t
(1)
(2)
-0.05∗∗
-0.07∗∗∗
(0.02)
-0.00
(0.01)
Crisisi,t × GFAi,t
(3)
(4)
(5)
(6)
(7)
(8)
-0.05∗∗
-0.07∗∗∗
-0.06∗∗∗
(0.03)
-0.03
(0.02)
-0.06∗∗∗
(0.02)
(0.02)
(0.03)
(0.02)
-0.05∗∗∗
(0.02)
-0.01
(0.01)
0.01
(0.01)
0.01
(0.01)
0.01
(0.01)
0.00
(0.01)
0.02∗
(0.01)
0.02∗
(0.01)
0.05∗
(0.02)
0.04∗∗
(0.02)
0.05∗∗
(0.02)
0.04∗
(0.02)
0.04∗∗
(0.02)
0.04∗∗
(0.02)
0.03∗
(0.02)
-0.07∗∗∗
(0.03)
-0.07∗∗∗
(0.03)
-0.07∗∗∗
(0.02)
-0.06∗∗
(0.03)
-0.06∗∗
(0.03)
0.05∗∗
(0.02)
0.05∗∗
(0.02)
0.05∗∗
(0.02)
0.05∗∗
(0.02)
Crisisi,t × GEDi,t
Crisisi,t × N IDi,t
Crisisi,t × Goldi,t
R2
Country fixed effects
Lagged Growth
Capital Flow Controls
Crisis in Sample
Observations
-0.02
(0.04)
0.090
✓
✓
0.096
✓
✓
0.278
✓
✓
✓
0.286
✓
✓
✓
0.307
✓
✓
✓
342
342
342
342
342
-0.02
(0.04)
0.125
✓
✓
✓
241
0.349
✓
✓
✓
✓
241
0.356
✓
✓
✓
✓
241
Notes: This table presents estimation results from altering Equation 6 to include additional sets of interaction terms. The dependent
variable is log GDP growth over the period t to t + 2. The independent variables are a financial crises indicator and its interaction with the
accumulation of gross foreign assets (GFA), gross exposure to foreign credit (GED), net capital inflows in year t (NID) and Gold Standard
adherence (Gold). The individual terms of each interaction are always included in the specification. The baseline capital flow variables
from the BoP, accumulated over t to t − 2, are included when indicated. All specifications control for country fixed effects and a two year
distributed lag of GDP growth. Standard errors in parentheses are dually clustered on country and year and *,**,*** indicates significance
at the 0.1, 0.05, 0.01 levels, respectively.
A19
Table A6.15: Capital outflow episodes and financial fragility
Episode
Reg. Sample
Flight
Flight
Retrenchment
Retrenchment
Level
All
High
Low
High
Low
N
363
43
33
36
28
Year
1931.2
1932.3
1934.0
1931.6
1932.1
GDP
Crises
∆2i,t+2
∆3i,t+3
∆4i,t+4
Ci,t
Ci,t+1
Ci,t+2
0.041
0.039
0.068
0.038
0.019
0.061
0.080
0.105
0.068
0.042
0.078
0.098
0.126
0.107
0.077
0.090
0.152
0.027
0.220
0.000
0.086
0.109
0.054
0.073
0.067
0.089
0.067
0.030
0.108
0.103
Notes: This table shows GDP growth and crisis likelihood following capital outflow episodes, combined with the levels at which they
occur. ’Flight’ (’retrenchment’) are observations where the current change in gross outflows lies one standard deviation above (below) its
’historic’ mean, defined as the rolling average of the previous three years. Levels are defined relative to the country-level median. For
comparison, the first row reports the same statistics for the regression sample.
A20
A7.
Channels
Table A7.16: Capital flows, the Gold Standard and capital account openness, full set of interactions
∆2 Yi,t+2
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
-0.04∗∗∗
-0.05∗∗∗
-0.05∗∗∗
-0.06∗∗∗
-0.07∗∗∗
-0.06∗∗∗
-0.06∗∗∗
-0.06∗∗∗
(0.01)
(0.02)
(0.01)
(0.02)
(0.02)
(0.02)
(0.02)
(0.02)
-0.06∗∗∗
(0.02)
No Goldi,t→t−2
0.04∗∗∗
(0.02)
0.04∗∗∗
(0.02)
0.04∗∗∗
(0.02)
Σ2j=0 Crediti,t− j × No Goldi,t→t−2
0.03∗
(0.02)
0.04∗
(0.02)
0.03∗∗
(0.02)
Σ2j=0 Crediti,t− j
Σ2j=0 Balancei,t− j
0.00
(0.01)
Σ2j=0 Debiti,t− j
0.01
(0.01)
-0.01
(0.01)
0.01
(0.01)
Σ2j=0 Balancei,t− j × No Goldi,t→t−2
0.00
(0.01)
-0.00
(0.01)
-0.01
(0.01)
Σ2j=0 Debiti,t− j × No Goldi,t→t−2
-0.00
(0.01)
Closed (< 100)i,t→t−2
0.05∗∗∗
(0.02)
0.05∗∗∗
(0.02)
0.05∗∗∗
(0.02)
Σ2j=0 Crediti,t− j × Closed (< 100)i,t→t−2
0.04∗∗
(0.02)
0.05∗∗∗
(0.02)
0.03∗
(0.02)
Σ2j=0 Balancei,t− j × Closed (< 100)i,t→t−2
-0.01∗∗∗
(0.00)
Σ2j=0 Debiti,t− j × Closed (< 100)i,t→t−2
0.02∗∗∗
(0.01)
Closed (< 67)i,t→t−2
0.04∗∗∗
(0.02)
0.04∗∗∗
(0.02)
0.04∗∗∗
(0.02)
Σ2j=0 Crediti,t− j × Closed (< 67)i,t→t−2
0.05∗∗
(0.02)
0.05∗∗
(0.02)
0.04∗∗
(0.02)
Σ2j=0 Balancei,t− j × Closed (< 67)i,t→t−2
-0.01
(0.01)
Σ2j=0 Debiti,t− j × Closed (< 67)i,t→t−2
R2
Country fixed effects
Lagged Growth
Observations
0.01
(0.01)
0.274
✓
✓
342
0.275
✓
✓
342
0.277
✓
✓
342
0.409
✓
✓
234
0.412
✓
✓
234
0.414
✓
✓
234
0.383
✓
✓
234
0.384
✓
✓
234
0.385
✓
✓
234
Notes: This table presents estimation results from interacting capital inflows with measures for Gold Standard adherence and capital
account openness. It shows the full set of interactions for all variables included in Table 5. The dependent variable is log GDP growth
from t to t + 2. Columns (1) to (3) interact BoP-flows with a dummy for not being on gold between t and t − 2. Columns (4) to (6) perform
a similar interaction with a dummy for the capital account being less than 100 percent open, based on the Quinn (2003) capital account
openness measure. Columns (7) to (9) repeat the specification for further reduced capital account openness (below 66%). All specifications
control for country fixed effects and lagged growth refers to a two year distributed lag of GDP growth. Standard errors in parentheses are
dually clustered on country and year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
A21
Table A7.17: Capital flows and official reserves
∆Foreign Asset Reservesi,t
(1)
Balancei,t
(2)
(3)
∆Gold Reservesi,t
(4)
-1.43∗∗∗
1.25∗∗∗
(0.45)
(0.42)
-1.25∗∗∗
(0.46)
Crediti,t
-1.55∗∗∗
(0.48)
(5)
(7)
(8)
(9)
0.51
(0.47)
0.37
(0.48)
0.56
(0.41)
0.49
(0.42)
0.97∗∗
(0.47)
Debiti,t
(6)
0.81∗
(0.44)
-1.02∗∗
(0.43)
-0.60
(0.48)
-3.23∗∗∗
(0.41)
Gold Balancei,t
-2.03∗∗∗
(0.49)
Gold Crediti,t
2.96∗∗∗
(0.46)
Gold Debiti,t
R2
Country fixed effects
Observations
-2.49∗∗∗
(0.48)
0.027
✓
389
0.020
✓
389
0.032
✓
389
0.024
✓
402
0.004
✓
402
0.018
✓
402
0.175
✓
382
0.057
✓
343
0.179
✓
343
Notes: This table presents estimation results for the relationship between one-year changes in the ratio of foreign asset reserves (columns
(1) to (3)) and gold reserves (columns (4) to (9)) held by the domestic central bank (both relative to notes in circulation), and capital
account flows from the Balance of Payments in year t. Columns (7) to (9) additionally include gold flows from the current account. All
specifications include country fixed effects. Standard errors, shown in parentheses, are dually clustered by country and year. *, **, and ***
indicate significance at the 0.1, 0.05, and 0.01 levels, respectively.
Table A7.18: Changes in official reserves and subsequent GDP growth
∆2 Yi,t+2
(1)
∆2 FA Reservesi,t−1
0.01
(0.01)
∆2 Gold Reservesi,t−1
(2)
∆3 Yi,t+3
∆4 Yi,t+4
(3)
(4)
(5)
(6)
(7)
(8)
(9)
0.01
(0.01)
0.01
(0.01)
0.02∗∗
0.02∗∗
0.02∗∗
0.03∗∗∗
0.02∗∗
(0.01)
(0.01)
(0.01)
(0.01)
(0.01)
0.02∗∗∗
(0.01)
0.00
(0.01)
0.00
(0.01)
0.00
(0.01)
0.00
(0.01)
-0.00
(0.01)
0.00
(0.01)
Σ2j=0 Crediti,t− j
-0.04∗∗∗
(0.01)
-0.06∗∗∗
(0.01)
-0.07∗∗∗
(0.01)
Σ2j=0 Debiti,t− j
0.00
(0.01)
0.01
(0.01)
0.01
(0.01)
R2
Country fixed effects
LDV
Observations
0.098
✓
✓
315
0.098
✓
✓
315
0.277
✓
✓
315
0.112
✓
✓
289
0.112
✓
✓
289
0.367
✓
✓
289
0.177
✓
✓
264
0.178
✓
✓
264
0.414
✓
✓
264
Notes: This table presents estimation results from regressing log GDP growth over horizons t to t + h on two-year changes in foreign
asset and gold reserves held by domestic central banks relative to notes in circulation. Columns (3), (6), and (9) additionally control for
the baseline gross capital flow variables. To allow for a direct comparison of coefficients, reserves are normalized prior to regression.
All specifications control for country fixed effects and a distributed lag of GDP growth, indicated by LDV (lagged dependent variable).
Standard errors, shown in parentheses, are dually clustered by country and year. *, **, and *** indicate significance at the 0.1, 0.05, and
0.01 levels, respectively.
A22
Table A7.19: Financial crises, foreign asset reserves and GDP growth
∆2 Yi,t+2
All
∆3 Yi,t+3
Ex. US
All
∆4 Yi,t+4
Ex. US
All
Ex. US
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
Crisisi,t
-0.05∗∗∗
(0.02)
-0.07∗∗∗
(0.02)
-0.06∗∗∗
(0.02)
-0.05∗
(0.03)
-0.08∗∗
(0.03)
-0.06∗∗∗
(0.02)
-0.04
(0.03)
-0.08∗∗
(0.03)
-0.06∗∗
(0.02)
FA Reservesi,t−1
-0.00
(0.03)
-0.01
(0.03)
-0.03
(0.03)
-0.01
(0.06)
-0.02
(0.06)
-0.05
(0.05)
-0.01
(0.08)
-0.04
(0.07)
-0.07
(0.07)
0.07∗
(0.04)
0.08∗∗
(0.03)
0.09
(0.06)
0.11∗
(0.05)
0.16∗∗
(0.08)
0.17∗∗
(0.07)
0.080
✓
✓
385
0.064
✓
✓
371
0.060
✓
✓
360
0.047
✓
✓
347
0.073
✓
✓
335
0.065
✓
✓
323
Crisesi,t × FA Reservesi,t−1
R2
Country fixed effects
Lagged Growth
Observations
0.076
✓
✓
385
0.056
✓
✓
360
0.063
✓
✓
335
Notes: This table presents estimation results from regressing log GDP growth over horizons t to t + h on the interaction of a financial crisis
indicator at time t with the level of foreign asset reserves at t − 1. Columns (3), (6), and (9) exclude the United States, an outlier in terms
of both recession severity and foreign asset reserves, from the sample. All specifications control for country fixed effects and include a
distributed lag of GDP growth, indicated by LDV (lagged dependent variable). Standard errors, shown in parentheses, are clustered by
both country and year. *, **, and *** indicate significance at the 0.1, 0.05, and 0.01 levels, respectively.
A23
A8.
Instrumental Variable Results
Table A8.20: US portfolio investments and changes in bank balance sheet size
∆Sizei,t
∆USPi,t
∆Liabilitiesi,t
∆Assetsi,t
(1)
Deposits
(2)
Debt
(3)
Reserves
(4)
Other
(5)
Loans
(6)
Bonds
(7)
Sec. ̸= Bonds
(8)
Other
(9)
0.97∗∗∗
(0.34)
0.68∗∗
(0.27)
0.14∗∗
(0.06)
0.09∗∗
(0.04)
0.04
(0.03)
0.45∗∗
(0.18)
0.28∗∗∗
(0.10)
0.08
(0.05)
0.16∗∗∗
(0.05)
✓
325
✓
325
✓
325
✓
325
✓
325
✓
325
✓
322
✓
325
✓
325
Country fixed effects
Observations
Notes: This table displays ’pass-through’ coefficients from changes in US portfolio investments (USP) to changes in domestic bank
balance sheets (BBS). Changes in bank balance sheets are further decomposed into the changes in deposits, debt instruments (cheques,
drafts, acceptances and re-discounts, interbank credit), reserves and ’other’ instruments on the liability side and loans, bonds, securities
(excluding bonds) and ’other’ instrument on the asset side. Changes are computed based on Fang et al. (2022) and Eren et al. (2023) as
described in Equation 9 and the individual components on the asset and liability side by construction sum to the total bank balance sheet
growth in column (1). All specifications include country fixed effects. Standard errors in parentheses are dually clustered on country and
year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
.2
.15
.3
.05
0
.2
.1
0
.1
.4
.04
.03
.02
Japan
.01
.25
France
.15
.2
United Kingdom
0
Germany
.3
.35
Figure A8.10: US portfolio investment relative to domestic bank balance sheet size
1923 1926 1929 1932 1935
1923 1926 1929 1932 1935
1923 1926 1929 1932 1935
1923 1926 1929 1932 1935
Notes: This figure plots the ratio of the United States portfolio position relative to domestic bank balance sheet size for Germany the
United Kingdom, Japan and France over time. Since this ratio is dependent on two time variant variables, changes in either one of them
might be driving the observed volume and trends. For Example: US portfolio investment volumes in Germany are consistently about 3 to
4 times as large as in the United Kingdom. The balance sheets of commercial banks in the UK, on the other hand, are three to four times
larger than the balance sheets of their German counterparts. This results in the ratio of US portfolio investments to domestic bank balance
sheets being about 10 times larger in Germany when compared to the United Kingdom.
A24
Figure A8.11: Capital exposure to the United States, 3 year cross-sectional GDP growth
Controlling for GDP Growth
.2
0
-.2
34
33
19
32
19
31
19
30
19
29
19
28
19
27
19
26
19
25
19
19
19
23
19
34
33
19
32
19
31
19
30
19
29
19
28
19
27
19
26
19
25
19
24
19
19
24
-.6
-.4
Cross Sectional Coefficient
.2
0
-.2
-.4
19
23
-.6
Cross Sectional Coefficient
.4
.4
No Controls
Notes: This figure shows coefficients for a cross-sectional regression of three year future GDP growth ∆3 Yi,t+3 on lagged exposure to
USP
US-portfolio investments BBS i,t−1 , for every year between 1923 and 1934. The left panel shows the unconditional relationship, the right
i,t−1
panel controls for contemporary GDP growth. Black lines represent 90% confidence intervals. The figure shows that a close financial
relationship with the United States was positively associated with growth in the early 1920sh, but that this link turned negative with the
onset of the Great Depression, before turning positive again afterwards.
Figure A8.12: First stage plot of the interaction instrument and capital inflows
With Controls
.5
-1
-1
-.5
0
Foreign Credit
0
-.5
Foreign Credit
.5
1
1
No Controls
-2
-1
0
Interaction Instrument
1
2
-2
-1
0
Interaction Instrument
1
2
Notes: This figure shows the first stage relationship between the Bartik-style instrument, constructed in Equation 10, and gross yearly
inflows at time t. Both variables are normalized and collapsed into equal sized bins. Each point represents the group specific mean.The
right panel includes country fixed effects and additionally controls for net capital inflows between t − 2 and t and GDP growth. Fitted
regression lines illustrate the positive correlation.
A25
Table A8.21: Predicting financial crises, instrumental variable results
Combined interwar crisis indicatori,t+1
OLS
(1)
Crediti,t
Reduced
(2)
0.05∗∗∗
(0.02)
OLS
(4)
0.12∗∗
(0.05)
0.09∗∗∗
(0.02)
0.04∗∗
(0.02)
Interaction IVi,t
Lagged Growth
Country fixed effects
Kleibergen-Paap Weak ID
Observations
IV
(3)
✓
✓
225
225
Reduced
(5)
IV
(6)
0.20∗∗∗
(0.07)
0.07∗∗∗
(0.03)
✓
✓
20.49
225
✓
✓
✓
✓
153
153
✓
✓
18.39
153
Notes: The table shows OLS, reduced form, and instrumented coefficients for a probit model for financial crises based on Equation 4. The
independent variables are yearly gross inflows, and the interaction instrument constructed in section 7. Lagged growth refers to a two
year distributed lag of GDP growth and country fixed effects are including when indicated. Standard errors in parentheses are clustered
on the country level and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
Table A8.22: Foreign capital inflows and changes in bank balance sheet size, instrumental variable results
∆Sizei,t
∆Liabilitiesi,t
∆Assetsi,t
(1)
Deposits
(2)
Debt
(3)
Reserves
(4)
Other
(5)
Loans
(6)
Bonds
(7)
Sec. ̸= Bonds
(8)
Other
(9)
Crediti,t
0.31∗∗∗
(0.08)
0.16∗∗∗
(0.06)
0.16∗∗∗
(0.04)
0.00
(0.02)
-0.01
(0.01)
0.23∗∗∗
(0.08)
0.18∗∗∗
(0.04)
-0.05
(0.03)
-0.03
(0.02)
Country fixed effects
Cragg-Donald Weak ID
Observations
✓
57.07
225
✓
57.07
225
✓
57.07
225
✓
57.07
225
✓
57.07
225
✓
57.07
225
✓
57.07
225
✓
57.07
225
✓
57.07
225
Crediti,t
0.14∗∗
(0.06)
0.06
(0.04)
0.07∗∗∗
(0.03)
-0.01
(0.01)
0.02∗∗∗
(0.00)
0.08∗∗∗
(0.03)
0.07∗∗
(0.03)
-0.01
(0.01)
0.01∗∗
(0.01)
R2
Country fixed effects
Observations
0.022
✓
225
0.008
✓
225
0.086
✓
225
0.003
✓
225
0.016
✓
225
0.021
✓
225
0.051
✓
225
0.002
✓
225
0.004
✓
225
IV estimated panel A:
OLS estimated panel B:
Notes: This table presents ’pass-through’ coefficients from gross foreign credit to changes in domestic bank balance sheets (BBS). Gross
foreign credit is instrumented with the interaction instrument constructed in section 7. Bank balance sheet changes are decomposed into
the changes in deposits, debt instruments (cheques, drafts, acceptances and re-discounts, interbank credit), reserves and ’other’ instruments on the liability side and loans, bonds, securities (excluding bonds) and ’other’ instrument on the asset side. Changes are computed
based on Fang et al. (2022) and Eren et al. (2023) as described in Equation 9. The individual components on the asset and liability side by
construction sum to the total bank balance sheet growth in column (1). All specifications include country fixed effects. Standard errors in
parentheses are clustered on the country level and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
A26
Table A8.23: Gross inflows and GDP dynamics, instrumental variable with dynamically adjusting exposure
∆2 Yi,t+2
OLS
(1)
Crediti,t
-0.02∗∗∗
(0.01)
∆3 Yi,t+3
IV
(3)
OLS
(4)
-0.04∗
(0.02)
-0.04∗∗∗
(0.01)
-0.02∗
(0.01)
Interaction IVi,t
Country fixed effects
Lagged Growth
Kleibergen-Paap Weak ID
Observations
Reduced
(2)
✓
✓
✓
✓
254
254
Reduced
(5)
∆4 Yi,t+4
IV
(6)
OLS
(7)
-0.08∗∗∗
(0.03)
-0.04∗∗∗
(0.01)
-0.04∗∗∗
(0.01)
✓
✓
20.91
254
✓
✓
✓
✓
254
254
Reduced
(8)
IV
(9)
-0.12∗∗∗
(0.03)
-0.05∗∗∗
(0.02)
✓
✓
20.91
254
✓
✓
✓
✓
245
245
✓
✓
16.05
245
Notes: This table presents OLS, reduced form and instrumented coefficients for a regression of log GDP growth between t and t + h on
gross foreign inflows at time t. The instrument deviates from the one described in Equation 10, as the exposure share to US-portfolio
investments is adjusted to be the mean of the previous two years, instead of being fixed to 1926. For the benefits and caveats associated
with this change, see text. The instrument is used to instrument gross inflows in columns (3), (6) and (9). All specifications control for
country fixed effects, a two year distributed lag of GDP growth and net capital flows. Standard errors in parentheses are dually clustered
on country and year and *,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
Table A8.24: Gross inflows and GDP dynamics, instrumental variable with normalized exposure and identical shocks
∆2 Yi,t+2
OLS
(1)
Crediti,t
-0.03∗∗
(0.01)
IV
(3)
OLS
(4)
-0.07∗∗∗
(0.02)
-0.04∗∗∗
(0.01)
-0.02∗∗∗
(0.01)
Interaction IVi,t
Country fixed effects
Lagged Growth
Kleibergen-Paap Weak ID
Observations
Reduced
(2)
∆3 Yi,t+3
✓
✓
✓
✓
225
225
Reduced
(5)
∆4 Yi,t+4
IV
(6)
OLS
(7)
-0.10∗∗∗
(0.02)
-0.04∗∗∗
(0.01)
-0.03∗∗∗
(0.01)
✓
✓
24.01
225
✓
✓
✓
✓
225
225
Reduced
(8)
IV
(9)
-0.15∗∗∗
(0.03)
-0.04∗∗∗
(0.01)
✓
✓
34.39
225
✓
✓
✓
✓
216
216
✓
✓
29.19
216
Notes: This table presents OLS, reduced form and instrumented coefficients for a regression of log GDP growth between t and t + h
on gross inflows at time t. The instrument deviates from the one described in Equation 10, to fulfill the formal requirements of a
Bartik-instrument. The exposure shares of 1926 are normalized to sum up to one, while the shock is the total change in the US portfolio
position and thus identical across countries. For the benefits and caveats associated with this change, see text. The alternative instrument
is used to instrument gross inflows in columns (3), (6) and (9). All specifications control for country fixed effects, a two year distributed
lag of GDP growth and net capital flows. Standard errors in parentheses are dually clustered on country and year and *,**,*** indicates
significance at the 0.1, 0.05, 0.01 levels, respectively.
A27
Figure A8.13: First stage plots of the GFC-measure and capital inflows
.75
0
-1.5
-.75
Capital inflows from t to t-2
.75
0
-.75
-1.5
Capital inflows from t to t-2
1.5
With Controls
1.5
No Controls
-1.5
-.75
0
GFC from t to t-2
.75
1.5
-1.5
-.75
0
.75
GFC if from t to t-2
1.5
Notes: This figure shows the first stage relationship between the GFC-measure (Global Financial Cycle), based on principal component
analysis, and the baseline variable for cumulative gross capital inflows between t − 2 to t. Both variables are expressed as normalized
three year sums and collapsed into 15 equal sized bins. Each point represents the group specific mean. The right panel includes country
fixed effects and additionally controls for net capital inflows between t − 2 and t and GDP growth. Fitted regression lines illustrate the
correlation.
Table A8.25: Gross foreign inflows and GDP dynamics, global financial cycle instrument
∆2 Yi,t+2
OLS
(1)
Σ2j=0 Crediti,t− j
-0.04∗∗∗
(0.01)
Σ2j=0 GFC−i,t− j
Country fixed effects
Lagged Growth
Net Capital Inflows
Kleibergen-Paap Weak ID
Observations
Reduced
(2)
∆3 Yi,t+3
IV
(3)
OLS
(4)
-0.14∗∗∗
(0.03)
-0.06∗∗∗
(0.02)
-0.05∗∗∗
(0.01)
✓
✓
✓
✓
✓
✓
323
323
Reduced
(5)
∆4 Yi,t+4
IV
(6)
OLS
(7)
-0.19∗∗∗
(0.04)
-0.06∗∗∗
(0.02)
-0.07∗∗∗
(0.01)
✓
✓
✓
18.01
323
✓
✓
✓
✓
✓
✓
296
296
Reduced
(8)
IV
(9)
-0.21∗∗∗
(0.05)
-0.07∗∗∗
(0.01)
✓
✓
✓
17.84
296
✓
✓
✓
✓
✓
✓
268
268
✓
✓
✓
17.59
268
Notes: This table presents OLS, reduced form and instrumented coefficients for a regression of log GDP growth between t and t + h
on cumulative gross capital inflows between t − 2 to t. In columns (3), (6) and (9) gross foreign inflows are instrumented with the
GFC-measure (Global Financial Cycle), based on principal component analysis. All specifications control for country fixed effects, a two
year distributed lag of GDP growth and net capital flows. Standard errors in parentheses are dually clustered on country and year and
*,**,*** indicates significance at the 0.1, 0.05, 0.01 levels, respectively.
A28