i i

“output” — 2023/11/20 — 15:34 — page 1 — #1

i i

Review of Economic Studies (2020) 0, 1–8 doi:10.1093/restud/rdaa002

© The Author(s) 2020. Published by Oxford University Press on behalf of The Review of Economic
Studies Limited.

Credit Allocation and

Macroeconomic Fluctuations
KARSTEN MÜLLER
National University of Singapore, Department of Finance.
E-mail: kmueller@nus.edu.sg
EMIL VERNER
MIT Sloan School of Management and NBER.
E-mail: everner@mit.edu

First version received June 2022; Editorial decision September 2022; Accepted July 2023 (Eds.)

We study the relationship between credit expansions, macroeconomic fluctuations,

and financial crises using a novel database on the sectoral distribution of private credit
for 117 countries since 1940. We document that, during credit booms, credit flows
disproportionately to the non-tradable sector. Credit expansions to the non-tradable
sector, in turn, systematically predict subsequent growth slowdowns and financial crises.
In contrast, credit expansions to the tradable sector are associated with sustained output
and productivity growth without a higher risk of a financial crisis. To understand these
patterns, we show that firms in the non-tradable sector tend to be smaller, more reliant
on loans secured by real estate, and more likely to default during crises. Our findings are
consistent with models in which credit booms to the non-tradable sector are driven by
easy financing conditions and amplified by collateral feedbacks, contributing to increased
financial fragility and a boom-bust cycle.

1. INTRODUCTION
Rapid expansions in private credit are often, but not always, followed by recessions
and financial crises (Schularick and Taylor, 2012; Jordà et al., 2013; Mian et al., 2017;
Greenwood et al., 2020). However, important questions about how private credit interacts
with the business cycle remain poorly understood. Why do some credit booms end
badly, while others do not? What are the mechanisms behind “good” from “bad” booms
(Gorton and Ordoñez, 2019)? Does it matter who takes on debt during these booms?
In this paper, we argue that the allocation of credit across sectors is important for
answering these questions. Our analysis is motivated by models of credit cycles with
sectoral heterogeneity and credit frictions (e.g., Schneider and Tornell, 2004; Reis, 2013;
Benigno and Fornaro, 2014; Kalantzis, 2015; Ozhan, 2020; Benigno et al., 2020). These
models distinguish between firms in the tradable and non-tradable sectors. Firms in the
non-tradable sector are assumed to be more financing constrained and more exposed

The editor in charge of this paper was Elias Papaioannou.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 2 — #2

i i

2 REVIEW OF ECONOMIC STUDIES

to feedbacks through collateral values and domestic demand linkages. This model set-
up yields two predictions about the link between the sectoral allocation of credit and
macroeconomic fluctuations. First, times of “easy credit” will lead to disproportionate
lending growth to firms in the non-tradable sector. Second, credit booms concentrated in
the non-tradable sector may lead to slower economic growth through increased financial
fragility. In contrast, lending to the tradable sector is more likely to coincide with strong
growth without increased financial fragility.
To examine the link between sectoral credit allocation and macroeconomic outcomes
empirically, we construct a novel database on private credit for 117 countries, starting
in 1940, by drawing on more than 600 sources. Existing datasets on credit distinguish,
at best, between firm and household lending. In contrast, our database covers up to 60
different industries. This allows us to differentiate between credit to the tradable and
non-tradable sectors, and key industries such as manufacturing, construction, and non-
tradable services. These new time series on credit by economic sector are consistent
with existing aggregate data on private credit. The data also cover a considerably
longer time span than other sources. We believe these data have many applications
in macroeconomics, finance, and international economics.1
Equipped with this database, we start by documenting that credit booms are
systematically associated with a reallocation of credit toward the non-tradable sector,
especially to the construction and real estate industries, alongside rapid growth in
household credit. Lending toward non-tradable firms and households accounts for about
70% of total lending growth during major credit booms. As a result, the share of credit
allocated to the non-tradable and household sectors rises in four out of five credit booms.
This reallocation rejects the view that credit booms are equally likely to increase leverage
in all sectors of the economy.
What explains this systematic reallocation of credit during booms? We document
that firms in the non-tradable sector are smaller and more reliant on debt secured by
real estate collateral relative to firms in the tradable sector. This suggests that non-
tradable firms are more financially constrained and more exposed to collateral feedbacks.
Therefore, the systematic reallocation of credit is consistent with an important role for
credit supply and asset price feedbacks in driving these kinds of booms. Further, credit to
the non-tradable sector is reinforced by demand feedbacks, as non-tradable sector firms
are more sensitive to booming domestic demand.
The allocation of credit during the boom predicts whether the boom ends in a
bust. While all credit booms coincide with strong output growth, only credit booms
concentrated toward non-tradable sector firms and households result in sharp growth
reversals. The magnitude of these growth reversals is sizeable. Five years after a credit
boom biased toward the non-tradable sector or households starts, real GDP is 5 percent
lower relative to a credit boom biased toward the tradable sector. As a result, there
is significant heterogeneity in the unconditional predictability of credit expansions for
future GDP growth. Expansion in credit to the non-tradable sector predicts subsequent
GDP growth slowdowns, defined as a significant decline in growth relative to the previous
trend. In contrast, a tradable sector credit expansion is associated with stable or, in

1. We discuss details of the data construction at length below and in the data appendix. Our
approach builds on best practices in the construction of national accounts used by the United Nations
(e.g., United Nations, 2009, 2018) and other data sources on private credit (e.g., Dembiermont et al.,
2013). We view our efforts as a reasonable starting point for constructing sectoral credit data in a
transparent and consistent way, which we plan to build on in the future.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 3 — #3

i i

MÜLLER AND VERNER CREDIT ALLOCATION 3

some specifications, higher growth in the medium run. Our analysis thus highlights that
heterogeneity within the corporate sector is important for understanding the aftermath
of credit expansions.
The patterns we document are robust to the inclusion of macroeconomic controls,
excluding the 2008 financial crisis, focusing solely on advanced or emerging markets,
controlling for year fixed effects or growth trends, and controlling for measures of the
riskiness of firm debt issuance based on the proxies used by Greenwood and Hanson
(2013). The results also hold after controlling for changes in sectoral value added, showing
that credit matters over and above variation in sectoral real activity. Further, while our
sectoral credit data generally do not systematically include bond market debt, various
tests incorporating information on bond issuance reinforce our findings.
Why does credit expansion to the non-tradable sector, but not to the tradable sector,
foreshadow lower future economic growth?2 Guided by theory, we present several pieces
of evidence that increased financial fragility and the risk of financial crises explain the
poor growth performance after non-tradable credit booms. At the outset, we emphasize
that causal identification of the exact mechanisms is challenging in such a broad and
long macro panel. Instead, our goal is to understand which theories are most consistent
with the empirical patterns.
First, credit expansion to the non-tradable sector is associated with a considerably
higher likelihood of a future systemic banking crisis. In contrast, lending to the tradable
sector, if anything, predicts a slightly lower probability of a banking crisis. The occurrence
of a banking crisis statistically accounts for the majority of the growth slowdown in the
aftermath of non-tradable credit expansions. Lending to the non-tradable sector also
falls dramatically after the onset of crises, indicating that this sector is more adversely
affected by credit contractions.
Second, loan losses during banking crises are concentrated in the non-tradable sector.
We collect data on non-performing loans by sector for ten major crisis episodes. When
non-performing loans reach their peak after banking crises, the share of non-performing
loans is 50% higher in the non-tradable compared to the tradable sector. Because credit
growth before crises is usually concentrated in non-tradable industries, the non-tradable
sector accounts for the majority of loan losses during banking sector meltdowns. In
contrast, the tradable and household sectors make up a much smaller fraction of losses.
Thus, defaults among firms in the non-tradable sector are key for understanding losses
during banking crises, as emphasized by the models of Schneider and Tornell (2004) and
Kalantzis (2015).
Third, non-tradable credit expansions are more strongly associated with real estate
price growth and subsequent busts. This pattern is consistent with greater financial
fragility from exposure to collateral feedbacks (Kiyotaki and Moore, 1997). Finally, non-
tradable credit expansions coincide with an appreciation of the real exchange rate and a
reallocation of labor and value added toward the non-tradable sector, suggesting rising
sectoral imbalances. At the same time, these booms predict lower future productivity
growth, consistent with the lower productivity in the non-tradable sector (Reis, 2013;
Benigno and Fornaro, 2014; Borio et al., 2016; Benigno et al., 2020). Lending to the

2. Given the established role of household credit expansions in predicting growth slowdowns
documented by Mian et al. (2017) and Jordà et al. (2020), among others, we focus most of our discussion
on the role of heterogeneity within the corporate sector. However, we always report results that control
for household credit, and, in the process, confirm the importance of household credit for predicting
growth slowdowns and crises in a larger sample than previous work.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 4 — #4

i i

4 REVIEW OF ECONOMIC STUDIES

tradable sector, on the other hand, is associated with higher productivity growth and a
stable real exchange rate.
This paper contributes to a growing literature on credit cycles. Previous studies find
that rapid growth in total private credit is associated with future growth slowdowns and
an increased risk of a financial crisis (Schularick and Taylor, 2012; Jordà et al., 2013).
Several studies examine the relative role of household and corporate credit during credit
expansions. Mian et al. (2017) find that credit expansion to households is associated with
a boom and subsequent bust in output, while there is less evidence for such a link for firm
credit (see also Drehmann et al., 2018; Jordà et al., 2020). In related work, Jordà et al.
(2016b) find that mortgage debt is associated with more severe recessions, compared to
non-mortgage debt, but that mortgage and non-mortgage debt have similar predictability
for financial crises. In contrast, Greenwood et al. (2020) find that credit booms coupled
with elevated asset prices, both in the household and corporate sectors, strongly predict
financial crises (see also Giroud and Mueller, 2020). Related studies find that elevated
credit market sentiment—proxied by times of increased lending to lower credit quality
firms—is correlated with credit expansions and predicts subsequent reversals in credit
market conditions and output (Greenwood and Hanson, 2013; López-Salido et al., 2017).
We provide several contributions to this literature. Our novel sectoral credit database
considerably extends existing datasets in terms of the sectors, countries, and time span
it covers. These data allow for new insights into the nature of credit booms that are
relevant for models featuring firm heterogeneity in financing constraints. Sufi and Taylor
(2021) argue that understanding financial crises requires investigating the boom that
precedes them. Our finding of a reallocation of credit toward non-tradable firms before
banking crises points to credit supply and collateral feedbacks as important factors. This
finding complements previous evidence on the importance of credit supply based on
credit spreads (Krishnamurthy and Muir, 2017; Mian et al., 2017) and debt issuance by
risky firms (Greenwood and Hanson, 2013).
Our new evidence on the importance of heterogeneity within the corporate sector
clarifies the mixed results about the link between corporate credit and macroeconomic
downturns. Beyond comparing household and firm debt, differentiating between different
types of firm credit is important. Our data allow us to explore the mechanisms for why
some credit booms end badly. Our new evidence on sectoral loan losses directly links
pockets of rapid firm credit growth to subsequent financial instability, supporting the
view that many financial crises are credit booms gone bust. In addition, our evidence
speaks to the tension between the literature emphasizing the benefits of credit for growth
(Levine, 2005) and studies linking credit booms to subsequent economic downturns.
Differentiating between different types of credit may not only matter for understanding
downturns, but also for longer-run growth outcomes.
Finally, our paper also contributes to the literature on capital inflows (Calvo et al.,
1996). Benigno et al. (2015) document that episodes of large capital inflows are associated
with booms and busts, along with a reallocation of labor out of manufacturing (see also
Tornell and Westermann, 2002; Schneider and Tornell, 2004). Diebold and Richter (2021)
document that much of the increase in credit-to-GDP has been financed by foreign capital
and that credit booms financed with capital inflows are likely to be followed by growth
slowdowns. Many of the credit booms we examine also stem from capital inflows.
The paper proceeds as follows. Section 2 describes our novel sectoral credit database
and presents new stylized facts about the evolution of credit markets around the world.
Section 3 discusses our conceptual framework for why credit expansion in certain sectors

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 5 — #5

i i

MÜLLER AND VERNER CREDIT ALLOCATION 5

may be linked to boom-bust cycles. Sections 4 to 6 present the main results and explore
mechanisms, and Section 7 provides concluding remarks.

2. SECTORAL CREDIT DATABASE: DATA AND METHODS

In this section, we outline the construction of our new sectoral credit database. We
address additional technical details and comparisons with other data sources in much
greater detail in a dedicated data appendix.3

2.1. Data Coverage

Existing datasets on private credit at best differentiate between household and firm
credit. These aggregated data, however, are not suitable for testing theories that link
sectoral credit expansions to economic fluctuations. We construct a new database on the
sectoral allocation of private credit covering the period 1940 to 2014. We assembled data
on credit by sector for 117 countries, which account for around 90% of world GDP today,
and include 53 advanced and 64 emerging economies. The number of sectors ranges from
2–60, with an average of 16. We also considerably extended the coverage of data on total
private credit, for which we cover up to 189 countries.
TABLE 1
Comparison with Existing Data Sources on Private Credit

Country- Country-sector-
Dataset Start Freq. Countries year obs. Sectors year obs.
Panel A: Sectoral credit data
Müller-Verner 1940 Y 117 5,436 Mean=16 89,019
Jordà et al. (2016a) 1870 Y 18 1,764 3 4,103
IMF GDD 1950 Y 83 1,871 2 3,703
BIS 1940 Q 43 1,220 2 2,417
Panel B: Total credit data
Müller-Verner 1910 Y 189 10,272 — 10,272
IMF IFS 1948 Y/Q/M 182 8,458 — 8,458
World Bank GFDD 1960 Y 187 7,745 — 7,745
IMF GDD 1950 Y 159 6,802 — 6,802
Monnet and Puy (2019) 1940 Q 46 2,936 — 2,936
BIS 1940 Q 43 2,020 — 2,020
Jordà et al. (2016a) 1870 Y 18 1,816 — 1,816

Notes: Panel A compares data that differentiate between different sectors of the economy (e.g., household vs.
firm credit). Panel B compares different sources of data on total credit to the private sector. WB GFDD stands
for the World Bank’s Global Financial Development Database (Cihák et al., 2013). BIS refers to the credit
to the non-financial sector statistics described in Dembiermont et al. (2013). IMF IFS and GDD refer to the
International Monetary Fund’s International Financial Statistics and Global Debt Database (Mbaye et al.,
2018), respectively. The data in Monnet and Puy (2019) is from historical paper editions of the IMF IFS.
Country-year obs. refers to the number of country-year observations covered by the datasets. Sectors refers
to the number of covered sectors; the mean refers to the average number of sectors in a country-year panel.
Country-sector-year obs. refers to country-sector-year observations. We count observations until 2014.

3. The sectoral credit database and the data appendix are available at
http://www.globalcreditproject.com.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 6 — #6

i i

6 REVIEW OF ECONOMIC STUDIES

Table 1 compares our database to existing datasets on private credit. Panel A

highlights the difference in our approach. The most disaggregated available data in Jordà
et al. (2016a) differentiates between household, firm, and mortgage credit for 18 advanced
economies. Our database contains a more detailed sectoral breakdown for many more
countries. It covers more than three times the country-year observations in Jordà et al.
(2016a) and more than four times the data on household and firm credit published by the
Bank for International Settlements (BIS). Because of the sectoral structure of our data,
it contains a total of 89,019 observations, orders of magnitude more than previous work.
Panel B shows how our database extends series on total credit to the private sector.
Here, we add long-run data starting in 1910 for a significant number of countries.

2.2. Data Sources

Most countries have collected sectoral credit data for several decades. However, historical
data are often not available in digitized form and are not reported on a harmonized
basis. We draw on hundreds of scattered sources to construct these time series. The
main sources are statistical publications and data appendices published by central banks
and statistical offices. A large share of the data was digitized for the first time from
PDF or paper documents. Many national authorities also shared previously unpublished
data with us. In the process, we discovered many untapped sources of total credit to the
private sector that allow us to extend existing time series.
We complement our newly collected data with existing time series from the BIS
(Dembiermont et al., 2013), Jordà et al. (2016a), the IMF’s International Financial
Statistics (IFS) and Global Debt Database (GDD, Mbaye et al., 2018), and additional
data from the print versions of the IFS digitized by Monnet and Puy (2019). These
existing sources track broad credit aggregates such as total private credit or household
credit for a subset of the countries we consider.

2.3. Concepts and Methods

We are interested in the sectoral distribution of outstanding credit to the private
sector. Ideally, the data should follow a harmonized definition of corporations and
households, economic sectors and industries, and coverage of debt instruments. In
practice, there are systematic differences in classifications across countries and time
that require adjustments. To harmonize data from a wide range of sources, we draw
on the metadata in historical publications and consulted with the national authorities
publishing information on sectoral credit.
The resulting dataset measures end-of-period outstanding claims of financial
institutions on the domestic private sector. In most countries, this definition mainly
covers loans, including foreign currency loans. We also include the bond exposures
recorded on financial institutions’ balance sheets wherever they are reported. In practice,
however, domestic credit is almost entirely accounted for by loans, while bonds are often
held by foreign financial institutions.
We try to cover the entire financial system wherever possible. In most countries,
the data predominantly measures credit extended by deposit-taking institutions such
as commercial banks, savings banks, credit unions, and other types of housing finance
companies. Comparisons with existing sources suggest that, on average, our numbers are
in line with the IMF IFS or BIS data on bank credit to the non-financial private sector.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 7 — #7

i i

MÜLLER AND VERNER CREDIT ALLOCATION 7

At times, we find somewhat larger values than the data in Jordà et al. (2016a), which
largely covers lending by different types of banks.
To classify different sectors of the economy, we follow the System of National Accounts
(SNA 2008) in differentiating between households and corporations (United Nations,
2009).
We classify industries based on the International Standard Industrial Classification of
All Economic Activities (ISIC), Revision 4 (United Nations, 2008). Most countries have
adopted this standard for reporting sectoral data, including on credit. In most countries,
we can differentiate between credit to the major “sections” in ISIC parlance (Agriculture,
Mining, Manufacturing, and so forth). The data generally capture credit to the (non-
bank) private sector. However, most data sources do not systematically differentiate
between lending to private and state-owned corporations; in principle, the data thus also
include lending to state-owned firms. We do not include direct lending to general or local
governments.
A key issue when dealing with time series data covering long time periods is how to
deal with level shifts (or “breaks”). The most important challenge is to understand if
such breaks arise because of actual economic changes (e.g., large-scale debt write-offs) or
because of changes in classification (e.g., in the types of financial institutions covered).
To address this issue, we coded country-specific classification changes based on a reading
of the metadata and additional methodological publications, as well as exchanges with
the national authorities. We adjusted breaks due to methodological changes using chain-
linking, following methods used in previous datasets on private credit (Dembiermont
et al., 2013; Monnet and Puy, 2019). To guarantee internal consistency of the data,
we rescale chain-linked time series to match an aggregate such as “total credit to non-
financial corporations” when needed, in line with the United Nations’ recommendation
on backcasting national accounts (United Nations, 2018).

2.4. Variable and Sample Construction

For the purpose of this paper, we construct sectoral credit aggregates that distinguish
between lending to households and a set of broad non-financial industries. Specifically,
we differentiate between credit to agriculture (ISIC Rev. 4 section A); manufacturing and
mining (sections B and C); construction and real estate (sections F and L); wholesale and
retail trade, accommodation, and food services (sections G and I); as well as transport
and communication (sections H and J). We further group together agriculture with
manufacturing and mining as the “tradable sector” and the other three industry groups
as the “non-tradable sector,” similar to other studies in international macroeconomics
(e.g., Kalantzis, 2015).
We construct a country-year panel dataset by merging the new credit data with
macroeconomic outcomes, house prices, and value added by sector. For our main analysis,
we restrict the sample to 75 countries with a population greater than one million in
2000 to avoid the results being influenced by large fluctuations in very small countries.
Appendix Table A.1 reports the countries and years used in our main analysis. The
sample includes broad coverage of both advanced and emerging market economies. We
winsorize variables at the 1% and 99% level to mitigate the influence of outliers, although
our results are similar without winsorizing. Table 2 reports summary statistics for key
variables.
To investigate the characteristics of different sectors, we use data on firm size from
the OECD’s Structural Business Statistics (SBS) and compute the share of firms with

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 8 — #8

i i

8 REVIEW OF ECONOMIC STUDIES

TABLE 2
Descriptive Statistics

Panel A: Summary statistics

N Mean Std. dev. 10th 90th
Real GDP growth (t–3,t) 1,890 15.71 10.35 3.89 28.68

∆3 dkit
Non-tradables 1,890 0.83 3.83 -2.92 5.16
Tradables 1,890 0.03 2.26 -2.55 2.57
Household 1,890 2.12 4.18 -1.63 7.58
Agriculture 1,890 0.02 0.73 -0.66 0.66
Manuf. and Mining 1,890 0.01 1.87 -2.14 2.08
Construction and RE 1,890 0.54 2.20 -1.32 3.01
Trade, Accomodation, Food 1,890 0.19 1.73 -1.58 2.03
Transport, Comm. 1,890 0.11 0.75 -0.55 0.84

Panel B: Correlation matrix

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

∆3 dkit
(1) Non-tradables 1
(2) Tradables 0.46 1
(3) Household 0.45 0.15 1
(4) Agriculture 0.21 0.64 0.15 1
(5) Manuf. and Mining 0.47 0.88 0.11 0.25 1
(6) Construction and RE 0.81 0.29 0.45 0.13 0.30 1
(7) Trade, Accom., Food 0.79 0.44 0.28 0.22 0.44 0.37 1
(8) Transport, Comm. 0.55 0.29 0.22 0.084 0.32 0.29 0.33 1

Notes: Panel A shows summary statistics for the main estimation sample. Panel B plots Pearson
correlation coefficients for three-year changes in the credit-to-GDP ratio ∆3 dkit for all sectors k used
in the analysis.

less than 10 employees for each industry. We also collect data on the type of collateral
posted in different sectors, which we could identify for five countries (Denmark, Latvia,
Switzerland, Taiwan, and the United States). These data come from the national central
banks, banking regulators, or Compustat (for the United States).
We use data on gross domestic product (GDP) in current national currency,
investment, consumption, population, inflation, and nominal US dollar exchange rates
from the World Bank’s World Development Indicators, Penn World Tables Version 9.1
(Feenstra et al., 2015), IMF IFS, GGDC (Inklaar et al., 2018), Jordà et al. (2016a),
Mitchell (1998), and the UC Davis Nominal GDP Historical Series. For a few countries,
we use data from national sources: Taiwan (National Statistics), the United States
(FRED), and Saudi Arabia (Saudi Arabian Monetary Authority). For labor and total
factor productivity, we use data from the Total Economy Database (TED). Data on

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 9 — #9

i i

MÜLLER AND VERNER CREDIT ALLOCATION 9

effective real exchange rates comes from the World Bank, BIS, and Bruegel (Darvas,
2012).
We construct data on sectoral value added and inflation from EU KLEMS, the Gronin-
gen Growth and Development Centre (GGDC) 10-sector database (Marcel Timmer,
2015), United Nations, UNIDO, OECD STAN, World Input-Output Database (WIOD),
and the Economic Commission for Latin America and the Caribbean (ECLAC). We
evaluate each source on a country-by-country basis and select the one that appears to be
of the highest quality. At times, we combine multiple sources by chain-linking individual
series.
We use data on the onset of systemic banking crises from Baron et al. (2021), who
classify banking crises with data on bank equity crashes and narrative information on the
occurrence of panics and widespread bank failures. For countries not covered by Baron
et al. (2021), we use data from Laeven and Valencia (2018). For robustness, we also use
banking crisis start dates from Reinhart and Rogoff (2009b). For house prices, we use data
from the BIS residential property price series, OECD, Dallas Fed International House
Price Database (Mack and Martı́nez-Garcı́a, 2011), and Jordà et al. (2016a). Finally, to
measure changes in firm borrowing in the bond market, we draw on gross bond issuance
data from SDC Platinum.

2.5. Stylized Facts About Private Credit Around the World

In this section, we present three stylized facts about long-term trends in credit markets
based on our new database. We start by revisiting facts about the amount of outstanding
private credit relative to GDP and then turn to the main novelty of the data: the sectoral
distribution of credit.

Fact #1: Credit/GDP has risen sharply over the past five decades. We
begin with a look at the long-run development of total private credit-to-GDP around
the world. The novelty of our data here is mainly the extension of long-run credit series
to the period before 1960. Figure 1a plots the average credit-to-GDP ratio for advanced
and emerging economies. This figure confirms the “hockey stick” pattern of rising private
debt in advanced economies documented by Schularick and Taylor (2012), but it also
reveals that the rise in credit is less pronounced in emerging economies.

Fact #2: Household debt has boomed globally, while firm credit has
stalled. The newly constructed data allows us to provide a first glimpse at sectoral
credit allocation over time using a large number of countries. Figure 1b plots averages
of household and firm credit-to-GDP over time. This shows that most of the growth in
credit-to-GDP since the early 1980s is accounted for by a rise in household debt. Relative
to GDP, the rise in lending to firms has been modest. This reinforces previous evidence
in Jordà et al. (2016b), who showed a similar pattern for a smaller sample of 17 advanced
economies.

Fact #3: Firm credit has shifted from tradable sectors to construction,
real estate, and other non-tradable sectors. It is a well-known phenomenon
that countries undergo structural change as they develop, away from primary sectors
toward manufacturing and then service sectors. One may expect to find similar trends

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 10 — #10

i i

10 REVIEW OF ECONOMIC STUDIES

Credit to GDP (in %)

100

90
Advanced economies
80

70

60

50

40

30

20
Emerging economies
10

0
1950 1960 1970 1980 1990 2000 2010

(a) By country group

Credit to GDP (in %)

Household credit
Firm credit
60

40

20

0
1950 1960 1970 1980 1990 2000 2010

(b) By sector

Figure 1
Private Credit-to-GDP (in %) by Country Group and by Sector, 1950-2014

Notes: Panel (a) shows the unweighted cross-country average of the ratio of total private credit-to-GDP.
The average is estimated on the full sample of 58 advanced and 127 emerging economies over the
period 1950-2014. Advanced economies refer to the World Bank’s 2019 classification of “high income
countries”, and emerging economies refers to all others. Panel (b) plots the unweighted cross-country
average of sectoral credit-to-GDP. The average is estimated on the full sample of 54 advanced and 76
emerging economies, 1950-2014.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 11 — #11

i i

MÜLLER AND VERNER CREDIT ALLOCATION 11

in corporate credit. At the same time, the finding of rising household debt may suggest
an increasing role of the housing sector, at least in advanced economies. Can we detect
complementary patterns in the composition of corporate financing?
Figure 2 plots the share of six subsectors in total corporate credit: agriculture;
mining and manufacturing; construction and real estate; trade, accommodation, and food
services; transport and communication; and other sectors. Consistent with structural
change in the credit market, the share of lending to agriculture and industry has declined,
particularly since around 1980. This trend appears in both advanced and emerging
economies.
The second major trend is that construction and real estate lending has come to
make up considerable shares of corporate loan portfolios. In advanced economies, the
share of construction credit in the 1950s was negligible. Today, this share has risen to
around 24 percent. While the housing boom of the 2000s has clearly played a role, the
share had already grown in the 1990s. Strikingly, a similar pattern also holds true in
emerging economies. In 1960, lending to industry and agriculture accounted for more
than 73 percent of corporate financing. Today, the ratio is closer to 26 percent. At
the same time, construction and real estate has increased from around 5 percent to 14
percent. Other services have also seen a substantial increase in their lending share. For
example, in advanced economies, other services have increased from around 18 percent
in 1960 to around 35 percent in recent years. Taken together, these findings suggest that
the financing of manufacturing, the activity perhaps most commonly associated with
commercial banking, has come to play a minor role for understanding modern credit
markets.

3. CONCEPTUAL FRAMEWORK
This section lays out a conceptual framework that motivates our empirical analysis.
We address the following questions. Which factors cause credit booms? What leads
credit booms to be concentrated in particular sectors of the economy? Does the sectoral
allocation of credit matter for whether a credit boom increases financial fragility and
triggers a subsequent output decline? We organize the discussion around two hypotheses
about credit expansions: the easy credit hypothesis and the productivity-enhancing credit
hypothesis. Given the prior evidence on household debt in credit cycles (Mian et al.,
2017; Jordà et al., 2020), we focus our discussion on heterogeneity within the corporate
sector.

3.1. Easy Credit Hypothesis

Credit supply expansion and credit allocation. The easy credit hypothesis starts
with an expansion in credit supply. Lenders provide cheaper credit and increase their
willingness to lend to risky borrowers. The expansion in credit supply can be driven
by a variety of factors, including loose monetary policy, rapid capital inflows, financial
deregulation, and optimism following a period of good fundamentals.
How does credit supply affect the allocation of credit across sectors in the economy?
Easy credit should particularly affect sectors that are more financing constrained, as
well as those more exposed to feedbacks through collateral values and their reliance on
domestic demand. Our main measure of sensitivity to changes in credit conditions is to
differentiate between non-tradable and tradable sectors using our sectoral credit data.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 12 — #12

i i

12 REVIEW OF ECONOMIC STUDIES

Sector share
% of firm credit
100

Other sectors
80

60 Transport/Communication

Trade, Accomm., Food

40
Construction/RE

20
Manufacturing/Mining
Agriculture
0
1960 1970 1980 1990 2000 2010

(a) Advanced economies

Sector share
% of firm credit
100

Other sectors
80
Transport/Communication

60
Trade, Accomm., Food

40
Construction/RE

20
Manufacturing/Mining

Agriculture
0
1960 1970 1980 1990 2000 2010

(b) Emerging economies

Figure 2
Sector Shares in Corporate Credit

Notes: This figure plots the average ratio of individual sectors in total corporate credit separately for
advanced and emerging economies. The plots are based on a sample of 46 advanced and 54 emerging
economies. “Other sectors” is the residual of total firm credit and the sectors we use in our main analysis.
This residual mainly comprises other (largely non-tradable) service sectors. Countries differ significantly
in the detail of credit data reported for service sectors. To maximize the number of countries for this
exercise, we group these together into “other sectors.”

Lending to the non-tradable sector is especially exposed to changes in credit

conditions for three reasons. First, firms in the non-tradable sector are likely to be more
financing constrained. To support this idea, Table 3 shows that the share of firms with
less than 10 employees is considerably higher in the non-tradable sector. Small firms
are often more financing constrained than large firms because they are more likely to
be opaque, bank-dependent, and have low net worth. This is consistent with a large

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 13 — #13

i i

MÜLLER AND VERNER CREDIT ALLOCATION 13

literature in international macroeconomics that assumes that non-tradable sector firms
are more financially constrained than firms in the tradable sector.4
Second, firms in the non-tradable sector are nearly twice as reliant on credit secured by
real estate (see Table 3). This implies that non-tradable sector firms are more sensitive to
asset price feedbacks through a collateral channel. Greater reliance on secured debt also
provides additional evidence that these firms are more financially constrained (Berger
et al., 2016; Luck and Santos, 2019; Benmelech et al., 2020; Rampini and Viswanathan,
2022). Firms in the non-tradable sector are particularly reliant on secured debt because
they are often small, risky, opaque, and low net worth, partially because they are limited
to serving domestic markets.
Third, non-tradable sector firms are more sensitive to feedbacks from domestic
demand. A credit boom that increases domestic demand will increase demand for both
tradable and non-tradable goods. While tradables can be imported, non-tradables must
be produced domestically. This leads to a further increase in output and, potentially,
credit in the non-tradable sector (Mian et al., 2020).5

Does the sectoral allocation of credit matter for financial fragility?. An expansion
of credit supply may result in a reallocation of credit toward firms in the non-tradable
sector. But does the allocation of credit across sectors matter for whether the boom
increases financial fragility?
The same factors that lead non-tradable sector firms to disproportionately benefit
from an expansion in credit can also explain why such credit booms increase financial
fragility and are more likely to end in a bust. More severe financing frictions in the
non-tradable sector imply a greater sensitivity to a reversal in credit supply following a
negative real or financial shock. Reliance on lending secured by real estate allows non-
tradable sector firms to lever up during the boom, but also exposes them to tightening
borrowing constraints and the possibility of fire sales in the bust (Kiyotaki and Moore,
1997).6 Furthermore, firms in the non-tradable sector are often less productive, so lending
to the non-tradable sector can shift resources to less productive firms that are more
likely to default, as in the models of Reis (2013), Benigno and Fornaro (2014), and Bleck
and Liu (2018). The higher fragility of non-tradable sector firms can lead to large-scale
defaults that cause a banking crisis, depressing credit supply and output. If borrowers and
lenders do not fully anticipate the downside risks during non-tradable credit booms, this
can lead to disappointed expectations following an increase in defaults, as in behavioral
models of credit cycles (Bordalo et al., 2018; Maxted, 2019).

4. See, for example, Tornell and Westermann (2002), Schneider and Tornell (2004), Reis (2013),
Kalantzis (2015), Bleck and Liu (2018), Brunnermeier and Reis (2019), and Ozhan (2020).
5. Ozhan (2020) refers to the financing constraint and demand mechanisms as the “banking” and
“trade” channels of sectoral reallocation. A distinct prediction for the relevance of financial frictions,
which we test below, is that non-tradable sector leverage (e.g., credit-to-output) rises during credit
expansions and predicts subsequent output slowdowns.
6. A related form of financial fragility due to high leverage and falling asset prices arises from
currency mismatch through foreign currency debt in the non-tradable sector, especially in emerging
markets (Mendoza, 2002; Schneider and Tornell, 2004; Kalantzis, 2015). However, the empirical patterns
we document below are broadly similar in advanced and emerging economies, suggesting that foreign
currency debt is not the only channel that can lead to financial fragility.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 14 — #14

i i

14 REVIEW OF ECONOMIC STUDIES

TABLE 3
Comparing Non-Tradable and Tradable Sector Characteristics

Tradable/Non-tradable Key industries

T NT NT - T Manuf. Constr./RE Food, Accomm.
Small firm share 0.79 0.90 0.12 0.78 0.91 0.86

Mortgage share 0.19 0.36 0.17 0.18 0.67 0.56

Labor productivity growth 5.03 2.65 -2.38 4.82 2.52 2.74

Total factor productivity growth 2.02 0.51 -1.51 2.19 -0.20 1.07

Notes: This table compares sectoral characteristics of non-tradable and tradable industries. Small firm
share is defined as the share of businesses with less than 10 employees, based on the OECD Structural
and Demographic Business Statistics. Mortgage share is the share of loans secured on real estate relative
to all outstanding loans based on data from five countries: Denmark, Latvia, Switzerland, Taiwan, and
the United States. For Denmark, we define use the ratio of lending by mortgage banks in each sector
relative to total lending by mortgage and commercial banks, using data for 2014-2020 from Danmarks
Nationalbank. For Latvia, we use the share of loans secured by mortgages using data for 2006-2012 from
the Financial and Capital Market Commission. For Switzerland, we use the share of mortgage lending in
each sector using data for 1997-2020 from the Swiss National Bank. For Taiwan, we compute the share
of lending for real estate purposes in each sector using data for 1997-2015 from the Central Bank of the
Republic of China (Taiwan). For the United States, we construct the weighted average ratio of mortgages
and other secured debt (dm) to total long-term debt (dltt) using Compustat. Labor productivity growth
is defined as the average yearly percentage growth in value added per engaged person in 2005 PPP USD,
calculated based on data from EU KLEMS, WIOD, and OECD STAN, as well as data on sectoral relative
prices from GGDC. The estimates are based on data from 39 countries. Total factor productivity growth
is from EU KLEMS and is based on data from 18 countries.

3.2. Productivity-enhancing Credit Hypothesis

Productivity and credit. Credit growth could also reflect higher anticipated produc-
tivity and output growth. In basic permanent income hypothesis models, households and
firms demand more credit to finance consumption and investment in response to higher
expected future income or productivity, so expansions in credit should be associated
with stronger future growth (e.g., Aguiar and Gopinath, 2007). In Coimbra and Rey
(2017), a positive productivity shock leads to an increase in credit without endangering
financial stability. In their model, “productivity driven leverage booms are not a concern
for financial stability in the same way that credit supply driven ones are.”
Credit growth could also drive sustained output growth. One example is a financial
reform that increases the ability of the financial sector to channel resources to productive
but constrained firms. The finance and growth literature treats credit depth as a marker
of financial development, so rising credit could contribute to stronger long-run growth
(Levine, 2005).7

Does the sectoral allocation of credit matter for growth?. Productivity-enhancing

credit growth could occur in all sectors, but it may be more likely when credit is
financing the tradable sector, especially manufacturing. Manufacturing has a high level
of productivity and has seen high productivity growth. Table 3 shows that annual labor
productivity growth has been over 2 percentage points higher in the tradable compared

7. For example, dynamic models with financial frictions predict that a decrease in financing
frictions leads to capital inflows and improved capital allocation across firms, which increases aggregate
productivity (Midrigan and Xu, 2014; Moll, 2014).

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 15 — #15

i i

MÜLLER AND VERNER CREDIT ALLOCATION 15

to the non-tradable sector. Growth in TFP has been 1.5 percentage points higher in
the tradable relative to the non-tradable sector. Moving resources to manufacturing may
have a positive effect on aggregate growth rates, making manufacturing an “engine of
growth” (Rodrik, 2012, 2016). Tradable sectors are more likely to learn about foreign
knowledge through trade and foreign competition, and productivity gains in the tradable
sector may be associated with positive spillovers to other firms in the economy (e.g.,
Benigno and Fornaro, 2014). The tradable sector also accounts for a disproportionate
share of investments in innovation, which can contribute to long-run growth (Benigno
et al., 2020). Lending growth biased toward tradables may thus capture times of strong
subsequent productivity and output growth without an elevated risk of a financial crisis.

4. THE ALLOCATION OF CREDIT DURING CREDIT BOOMS

In this section, we examine the dynamics of credit growth across sectors during credit
booms. We first discuss several prominent case studies and then turn to more systematic
evidence.

4.1. Case Studies

To motivate our empirical analysis, we begin by investigating two case studies of
prominent credit booms.8 The first is the case of Greece, Spain, and Portugal in the run-
up to the Eurozone Crisis. The peripheral countries of the Eurozone experienced a major
boom-bust cycle over the period 2000-2012. The creation of the European Monetary
Union eliminated currency risk, which led to a large reduction in country spreads and
large capital flows from core to peripheral economies (Baldwin and Giavazzi, 2015). These
capital inflows financed rapid loan growth.
Which sectors of the economy were financed by this credit expansion? Panels (a)
through (c) in Figure 3 reveal a large increase in lending to real estate firms, construction
firms, and households. In relative terms, lending to the real estate sector grew the fastest
in Portugal and Spain, while the absolute increase in debt was largest for the household
sector in all three countries. In contrast, credit to the manufacturing sector stagnated.
The lending boom was associated with house price booms and stagnant productivity
growth, as relatively unproductive firms in the non-tradable sector expanded at the
expense of the more productive firms in the tradable sector (Reis, 2013). The Global
Financial Crisis of 2008 led to a reversal of inflows, a sharp contraction in credit, falling
asset prices, severe recessions, and banking crises.
The second case is that of Japan in the 1980s. Japan experienced a rapid credit
boom in the second half of the 1980s, which culminated in a prolonged period of banking
sector distress and slow growth in the 1990s. The credit boom followed a period of
gradual financial deregulation and loose monetary policy (Cargill, 2000). The boom was
characterized by surging stock and urban real estate prices, which reinforced speculative
investment in housing by real estate finance companies (Ueda, 2000).
Panel (d) in Figure 3 shows that the Japanese credit boom was associated with
significant credit reallocation across sectors. Real estate and household credit increased
by over 50 percent between 1985 and 1990. Credit to the accommodation and food service
sectors also boomed. In contrast, manufacturing credit declined during this period.

8. Appendix B provides additional case study evidence for 18 episodes.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 16 — #16

i i

16 REVIEW OF ECONOMIC STUDIES

Credit to GDP Credit to GDP
Index (1999=100) Index (1999=100)
700 250

600
200
Real estate
500
150 Households
400 Construction
Real estate
Trade, Accomm., Food
300 100
Manufacturing
− Construction
200
Households 50
Trade, Accomm., Food
100
Manufacturing
0
1995 2000 2005 2010 2015 1995 2000 2005 2010 2015

(a) Eurozone Crisis: Spain (b) Eurozone Crisis: Portugal

Credit to GDP Credit to GDP
Index (2002=100) Index (1985=100)
400 Construction
200

Households
300 Households

150 Real estate

200
Accomm., Food
Trade, Accomm., Food Construction
100
100 Manuf., Mining

0 50 Manufacturing
1995 2000 2005 2010 2015 1980 1985 1990 1995 2000

(c) Eurozone Crisis: Greece (d) Japanese Financial Crisis

Figure 3
Case Studies: The Eurozone and Japanese Crises

Notes: Panels (a)-(c) plot the ratio of sectoral credit-to-GDP for construction (ISIC Rev. 4 section
F), real estate (L), trade/accommodation/food (G + I), manufacturing (C), and households in Spain,
Portugal, and Greece around the time of the Global Financial Crisis and Eurozone crisis. Values for Spain
and Portugal are indexed to 100 in 1999 (the year the euro was introduced), while Greece is indexed
to 100 in 2002, as construction credit data only start in that year. Panel (d) plots the ratio of sectoral
credit-to-GDP for construction (ISIC Rev. 4 section F), real estate (L), trade/accommodation/food (G
+ I), manufacturing (C), and households around the Japanese banking crisis of the early 1990s. The
areas shaded in gray mark years the countries were in a systemic banking crisis as defined by Laeven
and Valencia (2018).

4.2. Credit Booms and Credit Allocation: Systematic Evidence

We next turn to a more systematic investigation. We start by defining major credit
booms as periods when private credit-to-GDP expands rapidly relative to its previous
trend. To operationalize this definition, we first detrend total private credit-to-GDP
using the Hamilton (2018) filter with a horizon of four years. Then, we identify credit
booms as the first year when detrended total credit-to-GDP exceeds its country-specific
standard deviation.9 This captures periods when credit is particularly high relative to
a slow-moving trend. With this procedure, we obtain 113 credit boom episodes in our
sample.

9. The results are similar when we use an HP-filter or identify credit booms as periods when the
three-year expansion in credit-to-GDP is in its top quintile.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 17 — #17

i i

MÜLLER AND VERNER CREDIT ALLOCATION 17

Which sectors account for the increase in private credit during credit booms? Figure 4
presents an event study of the average cumulative increase in credit-to-GDP during major
booms and breaks this down by sectors. Panel (a) plots the contribution of individual
corporate sectors and the household sector to the total increase in private credit-to-
GDP, while panel (b) reports the cumulative change for the non-tradable, tradable, and
household sectors. Event time t = 0 refers to the year in which the boom starts. We de-
mean the change in credit-to-GDP for each sector within each country to abstract from
the longer-term structural trends in sectoral credit documented in section 2.
Table 4 shows substantial heterogeneity in the importance of different sectors for the
credit expansion during credit booms. The largest increase in absolute terms is accounted
for by household credit. This is followed by credit to construction and real estate and the
trade, accommodation, and food services sectors. These sectors account for roughly 70%
of the total increase in private credit. Thus, credit booms are largely a story of lending
to the real estate sector, other non-tradable sectors, and households. The outsized role
played by construction and real estate, other non-tradables, and households also stands
out in case studies of many prominent credit booms and crises (see Appendix B). Among
tradable sectors, manufacturing and mining represent the largest increase relative to
GDP, while the expansion in lending to agriculture is small.
Expansion in credit-to-GDP (in p.p.)

15
6
Cumlative change in credit-to-GDP

Household
10

4
Non-tradable
5

2
0

-5 -4 -3 -2 -1 0 1 2 3 4 5 Tradable
Year relative to crossing credit boom threshold
0
Households Constr, RE Trade, Accom., Food
Transport, Comm. Manuf., Mining Agriculture -5 -4 -3 -2 -1 0 1 2 3 4 5
Other sectors Year relative to crossing credit boom threshold

(a) Sectoral breakdown of aggregate credit(b) Non-tradable, tradable, and household sec-
expansion tors

Figure 4
The Allocation of Credit During Credit Booms

Notes: This figure plots an event study of the cumulative change in private credit-to-GDP around credit
booms, broken down by sectors. Panel (a) presents the disaggregated industries, and panel (b) shows the
non-tradable, tradable, and household sector aggregates. The credit boom events are defined as periods
of large deviations from a Hamilton (2018) filter with a horizon of four years. The change in credit-to-
GDP in each sector is demeaned at the country level to abstract from longer-term trends in credit over
time within countries. “Other sectors” is a residual category that includes services not included in the
remaining industries.

4.3. Which Characteristics Shape Credit Allocation During Booms?

Which characteristics shape the allocation of credit across corporate sectors during credit
booms? We investigate this question by estimating versions of the following specification

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 18 — #18

i i

18 REVIEW OF ECONOMIC STUDIES

in a country-sector-year panel:

∆3 ln(di,s,t ) = αi +β1 Boomi,t +β2 (Boomi,t ×High Characteristics )+ϵi,s,t , (4.1)

where ∆3 ln(di,s,t ) is the three-year growth in credit-to-GDP in country i and sector s,

Boomi,t is an indicator for when the credit boom is identified, and High Characteristics
is an indicator for a sector being above the median in non-tradability (the inverse of
exports-to-value added), the share of small firms, or the reliance on real estate collateral.
We estimate this for the five corporate sectors for which we have a broad and consistent
panel. We use the log change in credit to capture a sector’s sensitivity to a boom;
this ensures that the reallocation documented in Figure 4 is not solely the product of
differences in sectoral size.
Table 4 presents the estimates of equation (4.1). Credit booms are associated with
23% higher three-year sectoral credit growth compared to other periods. However, there
is important heterogeneity across sectors. During credit booms, the three-year growth
rate in credit is 7.1% higher in the non-tradable sector, 7.1% higher in sectors with a
high share of small firms, and 5.1% higher in sectors with a high mortgage share.
The estimates in Table 4 are robust to the inclusion of country-year and country-
industry fixed effects. Country-year fixed effects absorb aggregate shocks to countries
and can be viewed as a “difference-in-differences” estimate of the reallocation of credit
toward more constrained sectors during credit booms. Country-industry fixed effects
absorb country-specific trends in sectoral credit growth.
In sum, credit booms feature a large reallocation of credit toward the non-tradable
sector, and, related to this, industries that are more financially constrained and exposed
to collateral feedbacks. Given that financially constrained firms and those relying on
real estate collateral are particularly sensitive to a relaxation in financing conditions,
this points to an important role for credit supply expansion during credit booms. These
patterns are in line with the predictions of open-economy models of credit cycles discussed
in section 3.

5. CREDIT ALLOCATION AND BUSINESS CYCLES

Does the sectoral allocation of credit matter for whether a credit boom ends in a
subsequent bust? Existing studies show that credit booms predict growth slowdowns and
financial crises. The previous section documented that credit booms are associated with
a reallocation of credit toward non-tradable sector firms and households. This section
shows that these two facts are related: credit booms that feature reallocation toward
non-tradable firms and households are more likely to end in growth slowdowns.

5.1. Growth Around Major Credit Boom Episodes

We start with the sample of credit boom events constructed in the previous section.
We then divide these booms into two groups based on the sectoral allocation of credit.
Specifically, we define “tradable-biased” and “non-tradable-biased” booms, depending
on whether the change in the share of tradable credit, sTit = dTit /(dTit +dN T HH
it +dit ), over
the previous five years is positive or negative. We denote these booms as BoomTit
and BoomN T
it , respectively. We group households and non-tradables to obtain two

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 19 — #19

i i

MÜLLER AND VERNER CREDIT ALLOCATION 19

TABLE 4
Credit Booms and Credit Reallocation

Dependent var.: 100×∆3 ln(Di,s,t /GDPi,t )

(1) (2) (3) (4) (5) (6) (7)
Boomi,t 23.0∗∗ 18.8∗∗ 18.7∗∗ 19.9∗∗
(1.73) (2.57) (1.95) (2.42)
Boomi,t ×Non-tradables 7.08∗ 7.04∗∗
(3.15) (2.57)
Boomi,t ×HighSmallFirmShares 7.12∗∗ 6.96∗∗
(2.15) (1.83)
Boomi,t ×HighMortShares 5.10∗ 5.33∗
(2.32) (2.01)
Country FE ✓ ✓ ✓ ✓
Industry FE ✓ ✓ ✓ ✓
Country×Year ✓ ✓ ✓
Country×Industry FE ✓ ✓ ✓
Observations 10,529 10,529 10,526 10,529 10,526 10,529 10,526
# Countries 76 76 76 76 76 76 76
R2 0.11 0.11 0.59 0.11 0.59 0.11 0.59
Notes: This table presents estimates of equation (4.1) in a country-sector-year panel, where the dependent
variable is the log change (times 100) in sectoral credit-to-GDP over the previous three years. There are
five sectors: agriculture; manufacturing and mining; construction and real estate; wholesale and retail
trade, accommodation, and food services; and transport and communication. Non-tradable industries are:
construction and real estate; wholesale and retail trade, accommodation, and food services; and transport
and communication. Non-tradable industries are classified based on the inverse of the exports to value-
added ratio in the United States. Industries with a high small firm share are: agriculture; construction
and real estate; and transport and communication. High mortgage share industries are: agriculture;
construction and real estate; and wholesale and retail trade, accommodation, and food services. See
Table 3 for details on the industry characteristics. Driscoll and Kraay (1998) standard errors with six
lags in parentheses. +, * and ** denote significance at the 10%, 5% and 1% level.

disjoint sets of events.10 In total, we identify 25 tradable-biased booms and 88 non-

tradable-biased booms in our sample. The preponderance of non-tradable-biased booms
is consistent with the systematic credit reallocation toward non-tradables documented
in the previous section.
We estimate the average dynamics of real GDP for five years around these booms
relative to “normal” times using the following specification:

yt+h −yt−1 = αih +βTh BoomTit +βN

h NT h
T Boomit +ϵit+k , h = −5,...,5. (5.2)

The inclusion of country fixed effects, αi , allows for different trend growth rates across
countries. Figure 5 presents the sequence of estimates {β̂Th , β̂N
h }. During the boom phase
T
from event time t = −5 to t = 0, cumulative real GDP increases faster than during normal
times for both types of booms. Growth then diverges starting at the top of the boom in
t = 0 depending on the allocation of credit. Tradable-biased booms see real GDP plateau
about 4 percentage points higher after the boom relative to periods without a boom. In
contrast, non-tradable-biased booms see a sharp decline in growth that is statistically
significantly different from tradable-biased booms at the 5% level. From the peak in event
time 0, GDP declines by about 5% relative to non-boom periods. Thus, the allocation
of credit during clearly identified major credit booms helps distinguish whether these
booms are followed by major growth slowdowns.

10. Appendix Figure A.1 shows the results are similar when separating booms based on the non-
tradable credit share, excluding household debt.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 20 — #20

i i

20 REVIEW OF ECONOMIC STUDIES

10

Cumulative real GDP growth

relative to trend (t-1=0) 5

+
*
-5 *
*
-10
-5 -4 -3 -2 -1 0 1 2 3 4 5
Event time

Non-tradable-biased booms
Tradable-biased booms

Figure 5
Output Dynamics around Tradable and Non-tradable Biased Credit Booms
Notes: This figure plots results from estimating equation (5.2). Time zero is defined as the first year in
which the credit boom is identified. Tradable-biased (non-tradable-biased) credit booms are defined as
booms in which the share of tradable-sector credit (non-tradable and household sector credit) rises from
time t = −5 to t = 0. The union of BoomT it and Boomit
NT
thus comprises all identified credit booms.
Dashed lines represent 90% confidence intervals based on Driscoll and Kraay (1998) standard errors with
lag length ceiling(1.5(3+h)). +, * and ** indicate that the difference between the estimates, β̂Th − β̂N
h ,
T
is statistically significant at the 10%, 5% and 1% level, respectively.

5.2. Growth Following Sectoral Credit Expansions

Do sectoral credit expansions have differential unconditional predictive content for
business cycles? To answer this, we estimate the path of real GDP following innovations
in sectoral credit-to-GDP using Jordà (2005) local projections. The specification we
estimate is:

J
XX J
X
∆h yit+h =αih + k
βh,j ∆dkit−j + γh,j ∆yit−j +ϵit+h , h = 1,...,H, (5.3)
k∈K j=0 j=0

where ∆h yit+h is real GDP growth from year t to t+h, αih is a country fixed effect,
and ∆dkit is the change in sector k credit-to-GDP from t−1 to t. As is standard in the
local projection framework, we control for lags of the dependent variable. We choose a
conservative lag length of J = 5 based on the recommendation in Olea and Plagborg-
Møller (2020), who show that impulse responses estimated from lag-augmented local
projections are robust to highly persistent data, even for impulse responses at long
horizons. We examine a horizon of H = 10 years based on the evidence in the previous
section that credit expansions and subsequent busts often play out over longer periods.
Standard errors are computed using the methods in Driscoll and Kraay (1998) with a
lag length of ceiling(1.5·h), to allow for residual correlation within countries, as well

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 21 — #21

i i

MÜLLER AND VERNER CREDIT ALLOCATION 21

residual correlation across countries in proximate years. We also report standard errors
two-way clustered on country and year, which tend to be slightly more conservative in
our application.
Figure 6 presents the impulse responses of real GDP to innovations in non-tradable
sector credit, tradable sector credit, and household credit given by the estimated sequence
k } for k ∈ {N T,T,HH}. Panel (a) presents results from an estimation
of coefficients {β̂h,0
that includes the tradable and non-tradable corporate sectors, and panel (b) presents
results that add household credit to the specification. We emphasize that these impulse
responses are not necessarily causal, but provide a sense of the predicted dynamics of
GDP following innovations in sector k credit, holding fixed GDP growth and credit
growth in other sectors.
The left panel in Figure 6a reveals that an innovation in non-tradable sector credit-
to-GDP is associated with slower GDP growth after three to four years. The decline
persists for several years, leaving GDP below its initial trend. In terms of magnitudes,
a one percentage point innovation in non-tradable credit-to-GDP predicts 0.8% lower
cumulated GDP growth over the next five years. In contrast, the right panel in Figure 6a
shows that expansion in tradable sector credit is not associated with lower GDP growth.
The predictive relation is positive in the medium-term after five years. A one percentage
point innovation in tradable credit-to-GDP predicts 0.6% stronger cumulated growth
over the next five years and 2.1% cumulated over ten years.
Panel (b) adds household credit to the estimation of equation (5.3). Household credit-
to-GDP innovations are a strong predictor of lower GDP after three to four years. This
confirms the result in Mian et al. (2017) with a sample that is more than twice as large.
The patterns implied by the estimates on ∆dN it
T and ∆dT are similar to panel (a),
it
but slightly more muted. As non-tradable and household credit are relatively strongly
correlated (see Table 2), the estimates for non-tradable sector credit fall by about one-
third with the inclusion of household credit. This is consistent with non-tradable and
household credit capturing similar periods of credit expansions, which theory suggests
may be explained by similar exposure to easy credit conditions and to collateral and
demand feedbacks.
Table 5 presents an alternative regression approach to examining the relation between
credit expansions and GDP growth in the short and medium run. We estimate the
following regressions for h = 0,...,5:

∆3 yi,t+h = αih +βhN T ∆3 dN T T T HH HH

it +βh ∆3 dit +βh ∆3 dit +ϵit+h , (5.4)

where the left-hand-side is the change in log real GDP from year t−3+h to t+h, αih
is a country fixed effect, and ∆3 dkit is the three-year change in sector k credit-to-GDP.
We use the three-year change in credit-to-GDP based on the observation from Figure 4
that credit expands rapidly over three to four years during credit booms (see also Mian
et al., 2017).
Panel A in Table 5 presents the estimates of (5.4) for tradable and non-tradable
credit, and Panel B adds household credit. Non-tradable credit expansions are positively
correlated with GDP growth contemporaneously (column 1). In the medium run,
however, the sign reverses (columns 4-6). At the strongest horizon of h = 3, the estimate
in Panel B implies that a one standard deviation increase in ∆3 dN T is associated with
0.70 percentage points lower growth from t to t+3. The pattern for household credit is
similar, though household credit has a weaker contemporaneous correlation with growth
(column 1) and stronger negative predictability further into the future (columns 4-6). The

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 22 — #22

i i

22 REVIEW OF ECONOMIC STUDIES

Non−tradable sector credit Tradable sector credit

3.0 3.0

2.0 2.0
Real GDP response, %

Real GDP response, %

1.0 1.0

0.0 0.0

−1.0 −1.0

−2.0 −2.0
0 2 4 6 8 10 0 2 4 6 8 10
Years after innovation Years after innovation

(a) Non-tradable and Tradable Sector Credit

Non−tradable sector credit Tradable sector credit Household sector credit

3.0 3.0 3.0

2.0 2.0 2.0

Real GDP response, %

Real GDP response, %

Real GDP response, %

1.0 1.0 1.0

0.0 0.0 0.0

−1.0 −1.0 −1.0

−2.0 −2.0 −2.0

0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
Years after innovation Years after innovation Years after innovation

(b) Non-tradable, Tradable, and Household Sector Credit

Figure 6
Output Dynamics after Credit Expansions in Tradable, Non-Tradable, and Household Sectors
Notes: This figure presents local projection impulse responses of real GDP following innovations in
tradable sector credit, non-tradable sector credit, and household credit (all measured relative to GDP).
The impulse responses are based on estimation of (5.3). Panel (a) includes non-tradable and tradable firm
credit, while panel (b) presents results from the same specification that also includes household credit.
Dashed lines represent 95% confidence intervals computed using Driscoll and Kraay (1998) standard
errors, and dotted lines represent 95% confidence intervals from standard errors two-way clustered on
country and year.

estimate for the h = 3 horizon implies that a one standard deviation increase in ∆3 dHH
it
is associated with 1.60 percentage points lower growth from t to t+3. In contrast, an
expansion in tradable sector credit is associated with positive growth in both the short
and medium run, although the individual estimates are not statistically significant.

5.3. Additional Results and Robustness

This section presents additional results for the predictive relation between sectoral credit
expansions and subsequent real GDP growth. Appendix A presents a series of additional
robustness exercises. These show that the main results on sectoral credit expansion and
real GDP growth are robust to accounting for bond issuance and to including a range of

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 23 — #23

i i

MÜLLER AND VERNER CREDIT ALLOCATION 23

TABLE 5
Sectoral Credit Expansion and GDP Growth

Panel A: Non-tradable and tradable sector credit

Dependent var.: GDP growth over...
(1) (2) (3) (4) (5) (6)
∆3 dkit (t-3,t) (t-2,t+1) (t-1,t+2) (t,t+3) (t+1,t+4) (t+2,t+5)
Tradables 0.087 0.11 0.19 0.30 0.38 0.39
(0.15) (0.18) (0.18) (0.19) (0.24) (0.26)
Non-tradables 0.46** 0.15 -0.18+ -0.38** -0.47** -0.43**
(0.088) (0.11) (0.10) (0.11) (0.14) (0.16)
Observations 1,890 1,820 1,748 1,677 1,605 1,533
# Countries 75 75 75 75 75 75
R2 0.05 0.01 0.01 0.02 0.03 0.03

Panel B: Including household credit

Dependent var.: GDP growth over...
(1) (2) (3) (4) (5) (6)
∆3 dkit (t-3,t) (t-2,t+1) (t-1,t+2) (t,t+3) (t+1,t+4) (t+2,t+5)
Tradables 0.086 0.095 0.16 0.26 0.33 0.34
(0.14) (0.17) (0.17) (0.17) (0.20) (0.23)
Non-tradables 0.47** 0.21+ -0.045 -0.19* -0.23** -0.19*
(0.075) (0.11) (0.100) (0.079) (0.069) (0.091)
Households -0.0070 -0.11 -0.25** -0.39** -0.53** -0.55**
(0.090) (0.088) (0.075) (0.071) (0.10) (0.13)
Observations 1,890 1,820 1,748 1,677 1,605 1,533
# Countries 75 75 75 75 75 75
R2 0.05 0.01 0.02 0.05 0.08 0.08
Notes: This table presents the results from estimating (5.4). The dependent variable in column h is
the change in log real GDP (times 100) from year t−3+h to t+h. The right-hand-side variables,
∆3 dkit , are the changes in the credit/GDP ratio (in percentage points) for sector k from t−3 to t.
Driscoll and Kraay (1998) standard errors in parentheses with lag length ceiling(1.5(3+h)). +, * and
** denote significance at the 10%, 5% and 1% level.

additional controls. We also explore alternative sector classifications and show that the
results hold across various subsamples.

Sector size or sector leverage?. Credit booms often involve a reallocation of real
activity from the tradable to the non-tradable sector.11 Is slower growth after non-
tradable credit expansions merely driven by an increase in the size of the non-tradable
sector, or is it driven by an increase in sectoral leverage?
We use two approaches to address this question. First, Appendix Figure A.2a presents
results from estimating (5.3) with additional controls for the share of the non-tradable
and tradable sectors in value added, which hold constant any reallocation of output to
the non-tradable sector. Second, Appendix Figure A.2b presents estimates of impulse
responses from (5.3) where we replace sectoral credit-to-GDP with credit scaled by

11. See the discussion of Table 8 below, as well as Kalantzis (2015) and Mian et al. (2020).

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 24 — #24

i i

24 REVIEW OF ECONOMIC STUDIES

sectoral value added. Credit-to-value-added captures an increase in sectoral leverage.

Both approaches reveal that the increase in credit to the non-tradable sector, not just an
increase in sectoral real activity, matters for predicting future growth slowdowns. This is
consistent with models that emphasize differences in financing constraints across sectors.

Sectoral allocation and credit risk. Recent studies find that increased lending to
riskier firms in the economy is associated with a subsequent tightening in credit market
conditions and macroeconomic downturns (Greenwood and Hanson, 2013; López-Salido
et al., 2017). Are our sectoral credit expansion measures simply picking up variation
captured by existing credit risk measures?
To address this question, we construct two proxies for credit risk based on the
measures introduced by Greenwood and Hanson (2013) for the United States. The
first measure, ISS, is the average riskiness of firms with high debt issuance minus
the average riskiness of firms with low debt issuance, where riskiness is measured as
either the expected default probability or leverage. We construct the ISS measure for
an international panel using firm-level data from Worldscope following Brandao-Marques
et al. (2019). The second measure, HY S, is the share of bond issuance by high-yield
firms constructed by Kirti (2018).12 These measures are only available for approximately
one-third of the country-years in our baseline sample.
Table A.2 in the Appendix shows that these credit risk measures are positively
correlated with credit expansion in all sectors. However, Appendix Table A.3 (rows 15-
16) shows that controlling for firm credit risk has little impact on our results on GDP
growth. These results imply that the allocation of credit to non-tradables and households
contains distinct information over and above the credit risk measures. While credit risk
moves hand in hand with credit expansions, it is the sectoral allocation of credit in
particular that helps differentiate between booms that end badly and those that do not.

6. THE ROLE OF FINANCIAL FRAGILITY

Why are some types of credit expansions associated with economic slowdowns, while
others are not? One potential channel could be that risks to financial stability vary
with what credit is financing in the economy. This section explores the role of financial
crises, concentrated banking sector losses, house price reversals, and sectoral imbalances
in contributing to downturns in the aftermath of sectoral credit expansions.

6.1. Financial Crises

Models of financial crises with sectoral heterogeneity suggest that credit growth to
non-tradables and households can increase financial fragility, as these sectors are more
sensitive to expansions and reversals in credit supply and the price of assets used as
collateral (Mendoza, 2002; Schneider and Tornell, 2004; Kalantzis, 2015; Coimbra and
Rey, 2017; Ozhan, 2020). Because financial crises are associated with large costs in terms
of permanently lost output (Reinhart and Rogoff, 2009a), this may create a link between
sectoral credit expansions and future macroeconomic performance.
We start with a descriptive event-study analysis that examines how credit evolves
across sectors around the start of financial crises. Figure 7 plots the average yearly

12. Kirti (2018) has generously posted his international panel of high-yield share estimates on his
https://sites.google.com/site/divyakirti/webpage.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 25 — #25

i i

MÜLLER AND VERNER CREDIT ALLOCATION 25

change in sectoral credit-to-GDP for five years before and after a systemic banking
crisis, relative to non-crisis times. The sample includes 59 crises. Panel (a) shows that
non-tradable firm and household credit tend to expand more rapidly than the sample
average in the run-up to crises. Non-tradable sector credit expands more than twice as
rapidly relative to GDP as tradable sector credit, surpassing the growth of household
debt in the three years immediately before crises.
Panels (b) and (c) in Figure 7 decompose the broad firm sectors into five
industry groups. The expansion in credit to agriculture, manufacturing/mining, and
transport/communication is muted in the run-up to financial crises. In contrast, there
is a strong expansion in credit to trade, accommodation, and food services and to
construction and real estate. This evidence shows that the reallocation of credit during
credit booms identified in section 4 also occurs in the run-up to financial crises. For
individual case studies illustrating these patterns, see Appendix B.
Once the crisis occurs, credit to the non-tradable sector declines more compared to the
tradable sector. This may reflect that lending in non-tradable industries was “excessive”
before the crisis, leading to debt overhang (Myers, 1977). It is also consistent with models
where non-tradable sector firms are particularly exposed to contractions in credit supply
and tightening collateral constraints during crises (Ozhan, 2020), as crises are known to
disproportionately affect smaller firms and firms that are highly dependent on external
financing (Kroszner et al., 2007).
Next, we examine the predictability of financial crises based on expansions in credit
to different sectors. We run predictive panel regressions of the following form:
X
Crisisit+1 to it+h = αih + βkh ∆3 dkit +ϵit+h , (6.5)
k∈K

where Crisisit+1 to it+h is an indicator variable that equals one if country i experiences
the start of a systemic banking crisis between year t+1 and t+h, αih is a country fixed
effect, and ∆3 dkit the change in the credit-to-GDP ratio for sector k from year t−3 to
t. We thus estimate the predictive content of different credit expansions for cumulative
crisis probabilities.
Table 6 reports the results from estimating equation (6.5). Panel A examines the
predictive content of tradable, non-tradable, and household credit. Non-tradable and
household credit expansions predict an elevated probability of a financial crisis at the
one to four-year horizons. In terms of magnitudes, a two standard deviation higher three-
year change in non-tradable sector credit-to-GDP is associated with a 5% higher crisis
probability over the next year. This is sizeable relative to the unconditional probability of
a crisis of around 3%. For households, the magnitude is around 4%. In contrast, tradable
sector credit expansion predicts a slightly lower probability of a subsequent financial
crisis. The estimates on tradable sector credit are negative and mostly statistically
significant at the 10% level.13
Panel B shows the results for the individual corporate sectors. The estimates further
support the notion that banking crises tend to be preceded by credit expansions in
specific sectors of the economy. In particular, we find a strong role for lending to various
subsectors of the non-tradable sector: both lending to firms in the construction and real

13. Appendix A presents a series of sensitivity analyses and shows that the results on crisis
predictability are broadly robust to a range of additional controls, alternative specifications, alternative
crisis dates, and subsamples.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 26 — #26

i i

26 REVIEW OF ECONOMIC STUDIES

Change in credit/GDP
2.0
Household credit
1.5 Non−tradable credit
Tradable sector
1.0

0.5

0.0

−0.5

−1.0

−1.5

−2.0
−5 −4 −3 −2 −1 0 1 2 3 4 5

Years since banking crisis

(a) Tradable vs. Non-Tradable Sector

1.0 Agriculture
Manuf., mining
0.8

0.5

0.2

0.0

−0.2

−0.5

−0.8

−1.0
−5 −4 −3 −2 −1 0 1 2 3 4 5

Years since banking crisis

(b) Tradable Sector Industries

1.0 Construction/RE
Trade, Accomm., Food
0.8 Transport, Comm.

0.5

0.2

0.0

−0.2

−0.5

−0.8

−1.0
−5 −4 −3 −2 −1 0 1 2 3 4 5

Years since banking crisis

(c) Non-Tradable Sector Industries

Figure 7
Credit Dynamics around Systemic Banking Crises

Notes: This figure plots average annual percentage point changes in sectoral credit-to-GDP ratios around
59 systemic banking crises in 90 countries between 1951 and 2009. The horizontal axis represents the
number of years before and after a crisis. Crisis dates are from Baron et al. (2021), supplemented with
i i
dates from Laeven and Valencia (2018) for countries where they report no data.

i i
 i i

“output” — 2023/11/20 — 15:34 — page 27 — #27

i i

MÜLLER AND VERNER CREDIT ALLOCATION 27

TABLE 6
Sectoral Credit Expansions and Financial Crises

Panel A: Non-tradable, tradable, and household sector credit

Dependent variable: Crisis within...
∆3 dkit 1 year 2 years 3 years 4 years
Tradables -0.004+ -0.007* -0.007+ -0.004
(0.002) (0.004) (0.003) (0.004)
Non-tradables 0.006** 0.010** 0.011** 0.008+
(0.002) (0.002) (0.003) (0.004)
Household 0.004* 0.008* 0.010* 0.012**
(0.002) (0.003) (0.004) (0.004)
Observations 1,557 1,557 1,557 1,557
# Countries 72 72 72 72
# Crises 47 47 47 47
Mean Crisis Prob. 0.03 0.06 0.09 0.12
∆Prob. if 2 SD higher ∆3 dN
it
T 0.05 0.08 0.08 0.06
AUC 0.73 0.70 0.69 0.67
SE of AUC 0.04 0.03 0.03 0.02

Panel B: Individual corporate sectors

Dependent variable: Crisis within...
∆3 dkit 1 year 2 years 3 years 4 years
Agriculture -0.007 -0.008 -0.010 -0.010
(0.004) (0.009) (0.016) (0.017)
Manuf. and Mining -0.003 -0.008 -0.007 -0.003
(0.003) (0.005) (0.005) (0.006)
Construction and RE 0.008** 0.012** 0.011** 0.008
(0.003) (0.005) (0.004) (0.005)
Trade, Accomodation, Food 0.009** 0.020** 0.026** 0.028**
(0.003) (0.005) (0.008) (0.008)
Transport, Comm. -0.008 -0.018* -0.032** -0.044*
(0.006) (0.007) (0.010) (0.019)
Household 0.004* 0.007* 0.010** 0.012**
(0.002) (0.003) (0.003) (0.003)
Observations 1,557 1,557 1,557 1,557
# Countries 72 72 72 72
# Crises 47 47 47 47
Mean Crisis Prob. 0.03 0.06 0.09 0.12
AUC 0.75 0.74 0.72 0.71
SE of AUC 0.04 0.03 0.02 0.02

Notes: This table presents the results from estimating equation (6.5). In Panel A, we differentiate between
credit to the tradable, non-tradable, and household sectors. In Panel B, we use individual corporate
sectors. Crisis dates are from Baron et al. (2021), supplemented with dates from Laeven and Valencia
(2018) for countries not covered by Baron et al. (2021). Driscoll and Kraay (1998) standard errors with
lag length ceiling(1.5(3+h)) are in parentheses. +, * and ** denote significance at the 10%, 5% and 1%
level.

estate and in trade, accommodation, and food service sectors is associated with future
crises. At horizons of 2-4 years, these types of firm credit expansions have predictive
power that rivals or exceeds that of household credit. Credit to the primary sectors and
manufacturing have no predictability for banking crises.
We evaluate the performance of sectoral credit expansion in predicting crises through
the lens of the Area Under the Curve (AUC) statistic. The AUC is the integral of
a classifier’s true positive rate against its false positive rate for varying classification
thresholds (usually referred to as receiver operating characteristic curve, or ROC curve).
The AUC statistic measures a model’s ability to classify the data into crisis and non-
crisis periods. An AUC of 0.5 is thought of as containing classification ability no better
than a coin toss.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 28 — #28

i i

28 REVIEW OF ECONOMIC STUDIES

Sectoral NPL ratios in peak NPL year Sector shares in total peak NPLs
NPL to outstanding credit ratio Share in total NPLs
in percent in percent
Other
26
25
10%
20 Tradable
17
18%
15 Non−tradable
47%
11
10 8

5 25%

0
Non−tradable Tradable Other Households
Households

Figure 8
Financial Crises and Sectoral Loan Losses: Evidence from Ten Banking Crises
Notes: This figure documents sectoral differences in loan losses and the composition of non-performing
loans (NPLs) following ten systemic banking crises. The included crisis episodes (based on data
availability) are Mexico (1994), Thailand (1997), Indonesia (1997), Turkey (2000), Argentina (2001),
Italy (2008), Latvia (2008), Croatia (2008), Spain (2008), and Portugal (2008). Note that Laeven and
Valencia (2018) do not classify Croatia as experiencing a crisis in 2008, but Croatia did experience a
long-lasting recession following a period of rapid capital inflows and growth in corporate debt. The left
panel shows the median ratio of NPLs to outstanding loans separately for the non-tradable, tradable,
and household sectors in the peak NPL year. The right panel plots the median share of the individual
sectors in total non-performing loans in the year where the total NPL ratio reached its peak (within ten
years after each crisis).

The in-sample AUC in column 1 is 0.73, consistent with the informativeness of credit
expansion for predicting crises. In a rolling out-of-sample estimation, the corresponding
AUC is 0.75 (unreported). These AUC values are similar or slightly higher than the
AUCs from other studies using linear or logit models of crisis prediction. For example, in
a longer sample with fewer countries, Schularick and Taylor (2012) report AUCs of 0.67
to 0.72. Using only total private credit rather than sectoral credit on the same sample
as in Table 6 column 1, we obtain an AUC of 0.70, compared to 0.73 (panel A) or 0.75
(panel B) with sectoral credit measures.14
The increased likelihood of banking crises is central to understanding the slowdown in
real GDP growth in the aftermath of credit expansions toward non-tradable sectors. To
illustrate this, Appendix Figure A.6 presents local projection impulse responses of real
GDP growth to sectoral credit expansions separately for periods with a banking crisis
within the next three years and periods outside of banking crises. The real GDP response
to a non-tradable credit expansion is close to zero and insignificant outside of banking
crises, but large and significant for credit expansions that are followed by banking crises.
As an interesting contrast, the aftermath of household credit expansions is as severe
when excluding banking crises, consistent with theories emphasizing depressed household
demand from household debt overhang (Mian et al., 2021).

14. The AUCs are, however, considerably lower than those using more sophisticated machine
learning predictions (e.g. Fouliard et al., 2021).

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 29 — #29

i i

MÜLLER AND VERNER CREDIT ALLOCATION 29

6.2. Sectoral Defaults During Financial Crises
What ties sectoral credit expansions to a banking crisis that affects the economy
as a whole? In Figure 8, we provide evidence that sectoral losses are important for
understanding why the banking sector can end up in distress following credit expansion
to non-tradable sectors. To measure losses, we look at non-performing loans (NPLs),
which a few countries’ central banks or financial regulators report disaggregated by
sector, although usually only starting in the mid-2000s. We focus on ten financial crisis
episodes for which we were able to identify sectoral NPL data. See Figure 8 for the list
of episodes.
The left panel in Figure 8 plots the NPL rates of different sectors, and the right
panel shows a sector’s share in total NPLs. Banking crises tend to be followed by default
rates that are concentrated among firms in the non-tradable sector. Moreover, loans to
non-tradables and households account for nearly three-quarters of total bank NPLs post-
crisis, as shown in the right panel. Losses on loans to the non-tradable sector account
for nearly half of total NPLs in the aftermath of crises, while households account for a
quarter of NPLs.
These results have three important implications. First, they link the pockets of
rapid credit growth in the boom to banking sector distress in the crisis. This reinforces
the view that banking crises are often the consequence of loan losses following rapid
credit growth. Second, the evidence highlights that firms in the non-tradable sector are
particularly fragile following credit expansions, resulting in banking sector losses and
poor macroeconomic outcomes down the line. Third, the high rate of NPLs in the non-
tradable sector compared to the household sector suggests that banking sector distress is
important for explaining the slow growth after non-tradable sector credit booms, whereas
other channels such as weak household demand matter more for household credit booms.

6.3. House Price Booms and Busts

Credit expansions often coincide with strong growth in real estate prices. This connection
may be particularly strong for credit to non-tradables and households, as these sectors
rely heavily on loans collateralized by real estate (see Table 3). By relaxing collateral
constraints, increases in real estate prices can lead to increased borrowing by non-tradable
sector firms and households. In addition, an increase credit supply can itself boost real
estate prices (e.g., Greenwald and Guren, 2019). The aftermath of credit expansions, in
turn, often coincides with real estate price declines, generating feedback loops between
credit contraction and falling asset prices.
Table 7 investigates the dynamic relation between real estate price growth and sectoral
credit expansions. We estimate equation (5.4) with real house price growth as the
dependent variable. Column 1 shows that house price growth over t−3 to t is positively
correlated with credit expansions over the same three-year period. The correlation is
positive for all sectors and strongest for credit to the non-tradable sector.15
The subsequent columns in Table 7 reveal that non-tradable and household credit
predict a sizeable fall in future house price growth. A one standard deviation increase in
non-tradable credit expansion predicts 5.1 percentage points lower house price growth
from t to t+3. A one standard deviation increase in household credit expansion predicts

15. Appendix Figure A.7 confirms these patterns also hold in a local projection framework by
estimating (5.3) with log real house prices as the outcome variable.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 30 — #30

i i

30 REVIEW OF ECONOMIC STUDIES

TABLE 7
Sectoral Credit Expansions and House Price Growth

Dependent var.: Real house price growth over...

(1) (2) (3) (4) (5) (6)
∆3 dkit (t-3,t) (t-2,t+1) (t-1,t+2) (t,t+3) (t+1,t+4) (t+2,t+5)
Tradables 0.63 0.88+ 1.14∗ 1.06∗ 1.13∗∗ 0.85∗∗
(0.47) (0.51) (0.54) (0.50) (0.41) (0.26)
Non-tradables 0.95∗∗ 0.16 -0.68+ -1.16∗∗ -1.33∗∗ -1.04∗∗
(0.26) (0.34) (0.35) (0.32) (0.20) (0.19)
Households 0.49 0.16 -0.20 -0.64∗∗ -0.95∗∗ -0.90∗∗
(0.33) (0.34) (0.21) (0.15) (0.15) (0.20)
Observations 895 881 864 847 829 810
# Countries 42 42 42 42 42 42
R2 0.11 0.02 0.03 0.09 0.14 0.10
Notes: This table presents the results from estimating equation (5.4) with ∆3 ln(HP I)it+h (the
three-year change in log real house prices) as the dependent variable. All columns include country
fixed effects. Driscoll and Kraay (1998) standard errors in parentheses with lag length ceiling(1.5(3+
h)). +, * and ** denote significance at the 10%, 5% and 1% level.

3.0 percentage points lower house price growth over the same period. In contrast, tradable
credit expansions are associated with stronger future house price growth. This evidence
is consistent with heightened financial fragility through falling real estate prices following
expansions in non-tradable and household credit.

6.4. Sectoral Reallocation, Real Appreciation, and Productivity Dynamics

Credit expansion to the non-tradable sector can lead to sectoral imbalances and real
exchange rate overvaluation. This can increase financial fragility in multi-sector models
with asymmetric financing frictions (e.g., Schneider and Tornell, 2004; Kalantzis, 2015;
Rojas and Saffie, 2022). In models with sectoral heterogeneity in productivity dynamics
(Reis, 2013; Benigno and Fornaro, 2014; Benigno et al., 2020), sectoral imbalances further
lead to lower productivity. Low productivity growth is a direct source of low GDP growth
and can also increase the risk of a financial crisis (Gorton and Ordoñez, 2019). In contrast,
credit expansion to the tradable sector, often seen as an engine of growth, could finance
productivity improvements.
Table 8 investigates how sectoral credit expansions correlate with the sectoral
allocation of real activity and the real exchange rate. Columns 1 and 2 reveal that
non-tradable credit expansions coincide with a reallocation of real activity toward the
non-tradable sector, both in terms of output and employment. Credit expansion to the
non-tradable sector also correlates with a real exchange rate appreciation (column 3).16
While these patterns are not necessarily causal, they are consistent with the predictions of
multi-sector open economy models. Real appreciation could arise from strong domestic
demand that increases the scarcity of non-tradables (Mendoza, 2002; Schneider and
Tornell, 2004; Kalantzis, 2015; Mian et al., 2020). It could also be driven by misallocation
that leads to a higher cost per unit of produced non-tradable output (Reis, 2013). In
contrast, credit expansion to the tradable sector is not associated with significant sectoral
reallocation or real exchange rate appreciation.

16. Household credit expansions also contribute to a reallocation of real activity to non-tradables
and real exchange rate appreciation, consistent with a household demand channel of credit expansion
(Mian et al., 2020).

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 31 — #31

i i

MÜLLER AND VERNER CREDIT ALLOCATION 31

TABLE 8
Sectoral Credit Expansions, Sectoral Real Activity, and the Real Exchange Rate

(1)
 NT  (2)
 NT  (3)
∆3 dkit ∆3 ln YY T ∆3 ln EE T ∆3 ln(RER)
Tradables 0.29 -0.30 -0.32
(0.23) (0.19) (0.29)
Non-tradables 0.70∗∗ 0.43∗∗ 0.43∗
(0.16) (0.076) (0.20)
Households 0.41∗∗ 0.27∗∗ 0.31∗
(0.10) (0.058) (0.12)
Observations 1,638 846 1,793
# Countries 69 36 75
R2 0.09 0.14 0.03
Notes: This table presents regressions of changes in various macroeco-
nomic outcomes from t−3 to t on the expansion in tradable, non-tradable,
and household credit-to-GDP over the same period. The outcome
variables are the log of the non-tradable to tradable value added ratio
(column 1), the log of the non-tradable to tradable employment ratio
(column 2), and the log of the real effective exchange rate (column 3).
The real effective exchange rate is defined such that an increase signifies
real appreciation. All columns include country fixed effects. Driscoll and
Kraay (1998) standard errors in parentheses with lag length of 6. +, *
and ** denote significance at the 10%, 5% and 1% level.

TABLE 9
Sectoral Credit Expansions and Labor Productivity Growth

Dependent variable: Labor productivity growth over...

(1) (2) (3) (4) (5) (6)
∆3 dkit (t-3,t) (t-2,t+1) (t-1,t+2) (t,t+3) (t+1,t+4) (t+2,t+5)
Tradables 0.186+ 0.173∗ 0.208∗ 0.219+ 0.199 0.169
(0.093) (0.081) (0.086) (0.113) (0.150) (0.179)
Non-tradables 0.066 -0.072 -0.168 ∗ -0.147 ∗ -0.080 -0.009
(0.142) (0.125) (0.082) (0.067) (0.054) (0.059)
Households -0.147∗ -0.183∗∗ -0.226∗∗ -0.261∗∗ -0.312∗∗ -0.308∗∗
(0.061) (0.061) (0.057) (0.067) (0.076) (0.068)
Observations 1,451 1,451 1,451 1,451 1,451 1,451
# Countries 69 69 69 69 69 69
R2 0.01 0.01 0.03 0.03 0.03 0.03
Notes: This table presents the results from estimating equation (5.4) with the three-year change
in the log of labor productivity as the dependent variable. Driscoll and Kraay (1998) standard
errors in parentheses with lag length ceiling(1.5(3+h)). +, * and ** denote significance at the
10%, 5% and 1% level.

Table 9 examines whether credit expansions to different sectors predict differences

in future productivity. To do so, we replace the dependent variable in equation (5.4)
with the change in labor productivity, measured as the natural logarithm of output per
worker. The results in Table 9 show that credit expansions to the non-tradable sector
are systematically associated with lower productivity growth. The opposite is true for
lending to the tradable sector, which correlates with higher growth in labor productivity
in both the short and medium run.17

17. Appendix Table A.8 shows that the patterns are similar using total factor productivity growth
as the dependent variable.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 32 — #32

i i

32 REVIEW OF ECONOMIC STUDIES

7. CONCLUSION
There is increasing awareness that credit markets play a key role in macroeconomic
fluctuations. However, a lack of detailed, comparable cross-country data on credit
markets has left many questions about the relation between credit cycles and the
macroeconomy unanswered. By introducing a new worldwide database on sectoral credit,
this paper shows that heterogeneity in the allocation of credit across sectors—what credit
is used for—plays an important role for understanding linkages between the financial
sector and the real economy.
We document that credit expansions lead to disproportionate credit growth toward
non-tradable sector firms and households. This pattern is in line with theories in which
these sectors are more sensitive to relaxations in financing conditions and to feedbacks
through collateral values and domestic demand. The sectoral allocation of credit, in
turn, has considerable predictive power for the future path of GDP and the likelihood of
systemic banking crises. Credit growth to non-tradable industries predicts a boom-bust
pattern in output and elevated financial fragility. Credit to the tradable sector, on the
other hand, is less prone to large booms and is associated with higher future productivity
growth. Our evidence rejects the view that growth in private debt or leverage is uniformly
associated with subsequent downturns. It suggests that previous work, which could not
differentiate between different types of corporate credit, has missed an important margin
of heterogeneity.
While we are cautious about making welfare claims based on our reduced-
form evidence, these findings have interesting policy implications if taken at face
value. An ongoing policy debate has weighed whether financial regulation, including
macroprudential policy, should have a stronger focus on sectoral risks (Basel Commitee
on Banking Supervision, 2019b,a; European Banking Authority, 2020). Our results
suggest that such regulations could make sense, although there may be other concerns,
e.g., about political economy constraints (Müller, 2019). However, the debate about risks
in particular sectors has focused mainly on household debt and housing. We find that
lending to certain corporate sectors also matters.
Some caveats are in order. First, the importance of non-tradable and household
credit that we document here may be a more recent phenomenon. While we cover a
large proportion of economic downturns and crises since the 1950s, things may have
been different in the pre-World War II period. Second, while we point to a number
of potentially relevant sources of heterogeneity across industries, we cannot precisely
identify the exact underlying mechanisms. Third, the predictive patterns we document
in this paper are not necessarily causal. We hope that future work will find creative ways
to identify shocks to credit in different sectors, which could then be linked to economic
outcomes.

The data and code underlying this research is available on Zenodo at

https://doi.org/10.5281/zenodo.8347045.

Acknowledgment. We thank Elias Papaioannou (the editor), five anonymous referees, Matt Baron,
Thorsten Beck, Ben Bernanke, Ricardo Caballero, Mathias Drehmann, Stuart Fraser, Charles Grant,
Robin Greenwood, Sam Hanson, Daisuke Ikeda, Òscar Jordà, Şebnem Kalemli-Özcan, Enisse Kharroubi,
Paymon Khorrami, Deborah Lucas, Ernest Liu, Atif Mian, James Mitchell, Steven Ongena, Pascal
Paul, Romain Rancière, Veronica Rappoport, Oliver Rehbein, Björn Richter, Stephen Roper, Martin
Schneider, Moritz Schularick, Amir Sufi, Alan M. Taylor, David Thesmar, Adrien Verdelhan, Frank

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 33 — #33

i i

MÜLLER AND VERNER CREDIT ALLOCATION 33

Westermann, and participants at various seminars and conferences for comments and feedback. We
thank Maya Bidanda, Adamson Bryant, Niels Ruigrok, and Yevhenii Usenko for outstanding research
assistance. This paper incorporates parts of a draft previously circulated as “Sectoral Credit Around the
World, 1940-2014.” We gratefully acknowledge financial support from the Institute for New Economic
Thinking, the Governor’s Woods Foundation, the MIT Kritzman Gorman Fund, the National Science
Foundation (NSF Award 1949504), and a Singapore Ministry of Education Start-Up/PYP Research
Grant No. A-0003319-01-00. The new database on worldwide total and sectoral credit used in this paper
is available at http://www.globalcreditproject.com.

REFERENCES
Aguiar, M. and G. Gopinath (2007). Emerging Market Business Cycles: The Cycle Is
the Trend. Journal of Political Economy 115, 69–102.
Baldwin, R. and F. Giavazzi (2015). The Eurozone Crisis: A Consensus View of the
Causes and a Few Possible Solutions. VoxEU.
Baron, M., E. Verner, and W. Xiong (2021, 2). Banking Crises Without Panics. The
Quarterly Journal of Economics 136 (1), 51–113.
Basel Commitee on Banking Supervision (2019a). Guiding Principles for the
Operationalisation of a Sectoral Countercyclical Capital Buffer. Working paper.
Basel Commitee on Banking Supervision (2019b). Towards a Sectoral Application of the
Countercyclical Capital Buffer. Working Paper 36.
Benigno, G., N. Converse, and L. Fornaro (2015). Large Capital Inflows, Sectoral
Allocation, and Economic Performance. Journal of International Money and
Finance 55, 60–87.
Benigno, G. and L. Fornaro (2014, January). The Financial Resource Curse.
Scandinavian Journal of Economics 116 (1), 58–86.
Benigno, G., L. Fornaro, and M. Wolf (2020). The Global Financial Resource Curse.
FRB of New York Staff Report (915).
Benmelech, E., N. Kumar, and R. Rajan (2020). The Decline of Secured Debt. Technical
report, National Bureau of Economic Research.
Berger, A. N., W. S. Frame, and V. Ioannidou (2016). Reexamining the Empirical
Relation Between Loan Risk and Collateral: The Roles of Collateral Liquidity and
Types. Journal of Financial Intermediation 26, 28–46.
Bleck, A. and X. Liu (2018). Credit Expansion and Credit Misallocation. Journal of
Monetary Economics 94 (C), 27–40.
Bordalo, P., N. Gennaioli, and A. Shleifer (2018). Diagnostic Expectations and Credit
Cycles. The Journal of Finance 73 (1), 199–227.
Borio, C., E. Kharroubi, C. Upper, and F. Zampolli (2016, January). Labour Reallocation
and Productivity Dynamics: Financial Causes, Real Consequences. BIS Working
Papers 534.
Brandao-Marques, M. L., Q. Chen, C. Raddatz, J. Vandenbussche, and P. Xie (2019).
The Riskiness of Credit Allocation and Financial Stability.
Brunnermeier, M. and R. Reis (2019, September). A Crash Course on the Euro Crisis.
Working Paper 26229, National Bureau of Economic Research.
Calvo, G. A., L. Leiderman, and C. M. Reinhart (1996). Inflows of capital to developing
countries in the 1990s. Journal of economic perspectives 10 (2), 123–139.
Cargill, T. F. (2000). What Caused Japan’s Banking Crisis?, pp. 37–58. Boston, MA:
Springer US.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 34 — #34

i i

34 REVIEW OF ECONOMIC STUDIES

Cihák, M., A. Demirgüç-Kunt, E. Feyen, and R. Levine (2013). Financial Development

in 205 Economies, 1960 to 2010. Journal of Financial Perspectives 1 (2), 17–36.
Coimbra, N. and H. Rey (2017). Financial Cycles With Heterogeneous Intermediaries.
Technical report, National Bureau of Economic Research.
Darvas, Z. (2012, March). Real Effective Exchange Rates for 178 countries - A New
Database. Working Papers 716, Bruegel.
Dembiermont, C., M. Drehmann, and S. Muksakunratana (2013, March). How Much
Does the Private Sector Really Borrow - A New Database for Total Credit to the
Private Non-financial Sector. BIS Quarterly Review .
Diebold, L. and B. Richter (2021). When Two Become One: Foreign Capital and
Household Credit Expansion. Available at SSRN 3817595 .
Drehmann, M., M. Juselius, and A. Korinek (2018, April). Going with the Flows: New
Borrowing, Debt Service and the Transmission of Credit Booms. NBER Working
Papers 24549.
Driscoll, J. and A. Kraay (1998). Consistent Covariance Matrix Estimation With
Spatially Dependent Panel Data. The Review of Economics and Statistics 80 (4),
549–560.
European Banking Authority (2020). Final Guidelines on the Appropriate Subsets of
Sectoral Exposures in the Application of a Systemic Risk Buffer, EBA/GL/2020/13,,
30 September. Working paper.
Feenstra, R. C., R. Inklaar, and M. P. Timmer (2015, October). The Next Generation
of the Penn World Table. American Economic Review 105 (10), 3150–3182.
Fouliard, J., M. Howell, and H. Rey (2021). Answering the queen: Machine learning and
financial crises. Technical report, National Bureau of Economic Research.
Giroud, X. and H. Mueller (2020). Firm Leverage and Employment Dynamics. Technical
report.
Gorton, G. and G. Ordoñez (2019, 01). Good Booms, Bad Booms. Journal of the
European Economic Association 18 (2), 618–665.
Greenwald, D. and A. Guren (2019). Do Credit Conditions Move House Prices?
Unpublished manuscript, MIT .
Greenwood, R. and S. G. Hanson (2013). Issuer Quality and Corporate Bond Returns.
The Review of Financial Studies 26 (6), 1483–1525.
Greenwood, R., S. G. Hanson, A. Shleifer, and J. A. Sørensen (2020). Predictable
Financial Crises. Technical report.
Hamilton, J. D. (2018). Why You Should Never Use the Hodrick-Prescott Filter. The
Review of Economics and Statistics 100 (5), 831–843.
Inklaar, R., H. de Jong, J. Bolt, and J. van Zanden (2018). Rebasing ’Maddison’: New
Income Comparisons and the Shape of Long-run Economic Development. GGDC
Research Memorandum GD-174.
Jordà, Ò. (2005, March). Estimation and Inference of Impulse Responses by Local
Projections. American Economic Review 95 (1), 161–182.
Jordà, O., M. Kornejew, M. Schularick, and A. Taylor (2020, December). Zombies at
Large? Corporate Debt Overhang and the Macroeconomy. CEPR Discussion Papers
15518.
Jordà, Ò., M. Schularick, and A. M. Taylor (2013). Sovereigns Versus Banks: Credit,
Crises, and Consequences. Working Paper Series 2013-37.
Jordà, Ò., M. Schularick, and A. M. Taylor (2016a, September). Macrofinancial History
and the New Business Cycle Facts. In NBER Macroeconomics Annual 2016, Volume
31, NBER Chapters. National Bureau of Economic Research, Inc.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 35 — #35

i i

MÜLLER AND VERNER CREDIT ALLOCATION 35

Jordà, Ò., M. Schularick, and A. M. Taylor (2016b, 01). The Great Mortgaging: Housing
Finance, Crises and Business Cycles. Economic Policy 31 (85), 107–152.
Kalantzis, Y. (2015, 03). Financial Fragility in Small Open Economies: Firm Balance
Sheets and the Sectoral Structure. The Review of Economic Studies 82 (3), 1194–1222.
Kirti, D. (2018, January). Lending Standards and Output Growth. IMF Working Papers
2018/023, International Monetary Fund.
Kiyotaki, N. and J. Moore (1997, April). Credit Cycles. Journal of Political
Economy 105 (2), 211–48.
Krishnamurthy, A. and T. Muir (2017, September). How Credit Cycles across a Financial
Crisis. Working Paper 23850, National Bureau of Economic Research.
Kroszner, R. S., L. Laeven, and D. Klingebiel (2007). Banking Crises, Financial
Dependence, and Growth. Journal of Financial Economics 84 (1), 187 – 228.
Laeven, L. and F. Valencia (2018, September). Systemic Banking Crises Revisited. IMF
Working Papers 18/206.
Levine, R. (2005, 00). Finance and Growth: Theory and Evidence. In P. Aghion and
S. Durlauf (Eds.), Handbook of Economic Growth, Volume 1 of Handbook of Economic
Growth, Chapter 12, pp. 865–934. Elsevier.
López-Salido, D., J. C. Stein, and E. Zakrajšek (2017). Credit Market Sentiment and
the Business Cycle*. The Quarterly Journal of Economics 132 (3), 1373–1426.
Luck, S. and J. A. Santos (2019). The Valuation of Collateral in Bank Lending.
Mack, A. and E. Martı́nez-Garcı́a (2011). A Cross-Country Quarterly Database of Real
House Prices: A Methodological Note. Globalization Institute Working Papers 99,
Federal Reserve Bank of Dallas.
Marcel Timmer, Gaaitzen J. de Vries, K. d. V. (2015). Patterns of Structural Change in
Developing Countries, Chapter chapter4. Routledge.
Maxted, P. (2019). A macro-finance model with sentiment. Unpublished working paper .
Mbaye, S., M. Moreno Badia, and K. Chae (2018). Global Debt Database: Methodology
and Sources. IMF Working Papers 18/111.
Mendoza, E. G. (2002, January). Credit, Prices, and Crashes: Business Cycles with a
Sudden Stop, pp. 335–392. University of Chicago Press.
Mian, A., L. Straub, and A. Sufi (2021). Indebted demand. The Quarterly Journal of
Economics 136 (4), 2243–2307.
Mian, A., A. Sufi, and E. Verner (2020). How Does Credit Supply Expansion Affect
the Real Economy? The Productive Capacity and Household Demand Channels. The
Journal of Finance 75 (2), 949–994.
Mian, A. R., A. Sufi, and E. Verner (2017). Household Debt and Business Cycles
Worldwide. Quarterly Journal of Economics.
Midrigan, V. and D. Y. Xu (2014, February). Finance and Misallocation: Evidence from
Plant-Level Data. American Economic Review 104 (2), 422–58.
Mitchell, B. R. (1998). International Historical Statistics. Palgrave Macmillan UK.
Moll, B. (2014, October). Productivity Losses from Financial Frictions: Can Self-
Financing Undo Capital Misallocation? American Economic Review 104 (10),
3186–3221.
Monnet, E. and D. Puy (2019, September). One Ring to Rule Them All? New Evidence
on World Cycles. IMF Working Papers 19/202, International Monetary Fund.
Myers, S. C. (1977). Determinants of Corporate Borrowing. Journal of Financial
Economics 5 (2), 147–175.
Müller, K. (2019, December). Electoral Cycles in Macroprudential Regulation. ESRB
Working Paper Series 106, European Systemic Risk Board.

i i

i i
 i i

“output” — 2023/11/20 — 15:34 — page 36 — #36

i i

36 REVIEW OF ECONOMIC STUDIES

Olea, J. L. M. and M. Plagborg-Møller (2020). Local Projection Inference is Simpler

and More Robust Than You Think.
Ozhan, G. K. (2020). Financial Intermediation, Resource Allocation, and Macroeconomic
Interdependence. Journal of Monetary Economics 115, 265 – 278.
Rampini, A. A. and S. Viswanathan (2022). Collateral and secured debt. Working
Paper .
Reinhart, C. M. and K. S. Rogoff (2009a, May). The Aftermath of Financial Crises.
American Economic Review 99 (2), 466–72.
Reinhart, C. M. and K. S. Rogoff (2009b). This Time Is Different: Eight Centuries of
Financial Folly, Volume 1 of Economics Books. Princeton University Press.
Reis, R. (2013). The Portuguese Slump and Crash and the Euro Crisis. Brookings Papers
on Economic Activity, 143–193.
Rodrik, D. (2012, 11). Unconditional Convergence in Manufacturing. The Quarterly
Journal of Economics 128 (1), 165–204.
Rodrik, D. (2016). Premature deindustrialization. Journal of economic growth 21 (1),
1–33.
Rojas, E. and F. Saffie (2022). Non-homothetic sudden stops. Journal of International
Economics 139, 103680.
Schneider, M. and A. Tornell (2004). Balance Sheet Effects, Bailout Guarantees and
Financial Crises. The Review of Economic Studies 71 (3), 883–913.
Schularick, M. and A. M. Taylor (2012, April). Credit Booms Gone Bust: Monetary
Policy, Leverage Cycles, and Financial Crises, 1870-2008. American Economic
Review 102 (2), 1029–61.
Sufi, A. and A. M. Taylor (2021). Financial Crises: A Survey.
Tornell, A. and F. Westermann (2002, September). Boom-Bust Cycles in Middle Income
Countries: Facts and Explanation. Working Paper 9219, National Bureau of Economic
Research.
Ueda, K. (2000). Causes of Japan’s Banking Problems in the 1990s, pp. 59–81. Boston,
MA: Springer US.
United Nations (2008). International Standard Industrial Classification of All Economic
Activities, Revision 4. Technical report.
United Nations (2009). System of National Accounts 2008. Technical report.
United Nations (2018). Handbook on Backcasting. Technical report.

i i

i i