BIS Working Papers

No 1099
Who holds sovereign
debt and why it matters
by Xiang Fang, Bryan Hardy and Karen K Lewis

Monetary and Economic Department

May 2023

JEL classification: F34, G11, G15, F41.

Keywords: sovereign debt, banks and non-banks,

advanced economies and emerging markets.
BIS Working Papers are written by members of the Monetary and Economic
Department of the Bank for International Settlements, and from time to time by other
economists, and are published by the Bank. The papers are on subjects of topical
interest and are technical in character. The views expressed in them are those of their
authors and not necessarily the views of the BIS.

This publication is available on the BIS website (www.bis.org).

© Bank for International Settlements 2023. All rights reserved. Brief excerpts may be
reproduced or translated provided the source is stated.

ISSN 1020-0959 (print)

ISSN 1682-7678 (online)
 Who Holds Sovereign Debt and Why It Matters*

Xiang Fang Bryan Hardy

University of Hong Kong Bank for International Settlements

Karen K. Lewis
University of Pennsylvania, NBER, and CEPR

April 24, 2023

Abstract

This paper studies the impact of investor composition on the sovereign debt market. We construct
a data set of sovereign debt holdings by foreign and domestic bank, non-bank private, and official
investors for 95 countries over twenty years. Private non-bank investors absorb disproportionately
more sovereign debt supply than other investors. Moreover, non-bank investor demand is most
responsive to the yield. Counterfactual analysis of emerging market sovereigns shows a 10% increase
in debt leads to a 6.7% increase in costs, but an out-sized 9% increase if non-bank investors are absent.
We conclude that these sovereigns are vulnerable to losing non-bank investors.

JEL-Codes: F34, G11, G15, F41

Keywords: Sovereign debt, Banks and Non-banks, Advanced Economies and Emerging Markets

* We thank Maria Gelrud, Sophia Hua, and Max Yang for excellent research assistance. We also benefited from comments by
Karlye Dilts Stedman, Ester Faia, Gregor Matvos and participants at the Bank of International Settlements, Banque de France,
China International Finance Conference, Econometric Society North American Winter Meeting, Federal Reserve Board, Fed-
eral Reserve Bank of Dallas, Hong Kong University, INSEAD, London Business School, London School of Economics, the
Macro Finance Society, DebtCon5, EEA-ESSM Congress, the Midwest Finance Association, the National Bureau of Economic
Research, Society of Economic Dynamics, and the Wharton School of the University of Pennsylvania. Any errors or omis-
sions are our own. The views expressed here are those of the authors and not necessarily those of the Bank for International
Settlements. Contact: [email protected], [email protected], [email protected]
1 Introduction

The ability to issue debt is an important instrument at the government’s disposal. Sovereign

borrowing can help buffer the economy from the impact of adverse macroeconomic shocks.

Conversely, indebtedness can also make a country vulnerable to financial distress, as cri-

sis episodes have illustrated. Indeed, the sharp increase in fiscal expenditures and debt

issuance during the recent pandemic period, as well as concerns from the fallout of war,

has brought more urgency to understanding how a government can borrow. Answering

this question requires knowledge of who invests in sovereign debt and how these investors

impact borrowing costs. Therefore, this paper provides an analysis of who holds sovereign

debt and what this investor composition implies for governments’ borrowing costs.

We begin by documenting who holds government debt around the world and establish-

ing some new empirical regularities. For this purpose, we first assemble a dataset that dis-

tinguishes the holders of each country’s sovereign debt into investor groups that have been

highlighted in the literature. That is, we disaggregate debt held by foreign and domestic

investors and into three subgroups within those categories: private banks, other private in-

vestors that we term ”non-banks”, and official creditors consisting largely of central banks

and international organizations like the World Bank.1 Assembling these data series provides

1744 country-year observations for which we can decompose the holders of debt, spanning

95 countries over 1991-2018.2

Figure 1 shows both the growing importance of government debt as well as how the in-

vestor base varies in our data set. Specifically, Panel (a) shows that aggregate government

debt as a proportion of GDP has increased to the highest levels in recent history, spurred on

by the Covid-19 pandemic for both for advanced and emerging economies. Furthermore,

the composition of investor shares has changed over time. For example, as Panel (b) shows,

the share of domestic versus foreign investors has evolved with marked differences between

advanced economies (AEs) and emerging markets (EMs). The share of aggregate debt hold-
1 This decomposition follows that of Arslanalp and Tsuda (2012) and Arslanalp and Tsuda (2014).
2A partial decomposition is available for 151 countries.

1
ings by domestic investors has decreased for AEs while that same share has increased for

EMs. The share of debt held by non-bank private investors, shown in Panel (c), shows more

short-term fluctuations.
Figure 1: Trends in General Government Debt

100 0.8 0.8

75 0.6 0.6

50 0.4 0.4

25 0.2 0.2

0 0.0 0.0
03 06 09 12 15 18 21 00 03 06 09 12 15 18 00 03 06 09 12 15 18
AE EMDC AE (excl US) US EM (excl CN) CN

(a) Debt/GDP (b) Share Held Domestically (c) Share Held by Non-banks

Note: Panel (a) of this figure plots the time-series of general government debt-to-GDP ratio for the advanced economies,
emerging market and developing countries from IMF WEO. Panel (b) plots the share of general government debt held by
domestic investors by country group. Panel (c) plots the share of general government debt held by non-banks. Panels (b)
and (c) consist of balanced samples of 15 AEs including the US and 15 EMs including China (CN).

In this paper, we use these data to document how increases in sovereign debt are ab-

sorbed by the six different investor groups (foreign vs domestic and bank vs non-bank

vs official). Strikingly, we find that non-bank private investors increase their holdings of

sovereign debt at significantly higher rates than any other group, including private banks.

Furthermore, this absorption rate is greater than proportional to their average holdings.

Across all countries for instance, when there is an increase in debt, 69% of the increase is

allocated to non-bank investors, even though they make up only 46% of holdings on aver-

age. Moreover, for increases in foreign-held debt, 75% of the increase is due to non-bank

investors, even though they comprise only 42% of all foreign holdings. By contrast, banks

absorb less than their average holdings of sovereign debt: they hold 28% of the debt on av-

erage, but only take up 20% of new debt on the margin. This pattern holds for increases in

foreign and domestically held debt as well as in subsamples of AEs and EMs. Furthermore,

the general results are robust to accounting for currency valuation effects on foreign-held

2
debt. Thus, when the supply of government debt increases, private non-bank investors play

a significantly larger role than other investors in holding this expansion.

These aggregate results highlight the importance of non-bank private investors, a large

and heterogeneous group. To disentangle the behavior of different investors within this

group, we turn to more granular data sets. We begin by using data on Euro Area investors

to disaggregate the foreign non-bank investor group into non-financial corporations, pen-

sions and insurance companies, households, and a category of other financial institutions,

largely representing investment funds. Our analysis indicates that within this group, finan-

cial institutions such as investment funds drive the large response of non-banks. We then

repeat this analysis for domestic investors who hold US Treasuries and UK Gilts, finding

again that non-banks are the most responsive set of investors, and within domestic non-

banks that funds are the key players.

Given these findings for who holds sovereign debt, we next consider why this composi-

tion matters to borrowers. To answer this question, we estimate a system of funding equa-

tions representing the willingness to extend credit. Since this willingness to fund depends

endogenously on the interest paid by the government, we require an instrument for these

borrowing costs. For this purpose, we exploit the fact that our data set provides the full

breakdown of sovereign debt by investor type for each country. As shown by Koijen and

Yogo (2020), such a market clearing condition can be used to provide an instrument that

is uncorrelated with unobserved latent factors that drive investors’ willingness to extend

credit. Accordingly, we use this concept to construct an instrument for the sovereign yield.

Our estimates provide striking results that again point to the importance of non-bank

investors. Specifically, for both Emerging Markets and Advanced Economies, non-bank

investors are the creditor group most responsive to changes in yields. Moreover, investors

of emerging market sovereign debt generally dislike inflation and prefer income growth

from the borrowing country.

While these results provide measures of the sovereign’s view of each creditor group’s

willingness to fund, investors will perceive the returns on these investments differently for

3
two main reasons. First, while governments issue debt and pay interest over time, investors

focus upon holding returns per period as measured in the secondary market. Second, since

governments primarily issue debt in local currency, foreign investors who hold this debt

will incur currency risk in addition to the holding returns.

Although the aggregate data do not contain the identity nor portfolio of holdings by

these investor groups, we show that under some additional conditions we can provide in-

vestor demand estimates similar to those in Koijen and Yogo (2020) and Jiang, Richmond,

and Zhang (2022). First, we focus only on Emerging Markets because the identity of domes-

tic versus foreign investor groups is likely to be more comingled for Advanced Economies.

As a second condition, foreign investors are assumed to be US dollar-based. Importantly,

as shown by Koijen and Yogo (2020), the demand system under these assumptions allow

for substitution across debt issued by different countries. This estimation shows that the

elasticity of non-bank investors is again higher than that of banks, and that the elasticity of

foreign non-bank investors is the greatest.

To assess what these estimates imply for the sovereign borrower, we ask how much the

financing cost would increase for a hypothetical debt increase. Our estimates show that a

10% increase in debt corresponds to a 6.7% increase in yield for the average EM borrower.

However, if non-bank investors are not present and borrowers must borrow from banks,

the same 10% increase in debt corresponds to a substantially higher 9.1% increase in yield.

Thus, EM sovereigns appear highly exposed to the availability of non-bank investor fund-

ing. We also study the trade-off between borrowing costs and characteristics. For instance,

if inflation increases by 1%, the required yield for the average EM would increase by 2.5%.

The structure of the paper is as follows. We next provide a brief review of the litera-

ture. Section 2 describes the data and some basic stylized facts, including a decomposition

that highlights how much government debt each investor group absorbs on the margin. It

also reports the same decomposition using disaggregated non-bank investor holdings in

the Euro Area, the US Treasuries, and the UK Gilts. Section 3 sets up a framework to ex-

plore funding and investor demand in the sovereign debt market and reports estimates of

4
this demand by investor groups. Section 4 combines the country-level investor demand

identification with the marginal absorption estimates to develop a counterfactual analysis.

Concluding remarks follow.

Related Literature Since our paper studies investor behavior of sovereign debt holdings,

it relates to a number of different strands of research in macro-finance. First, it contributes

to a growing literature that uses the demand system approach to asset pricing introduced

by Koijen and Yogo (2019) and applied in domestic and international financial markets as

in Koijen and Yogo (2020), Koijen, Richmond, and Yogo (2020), Jiang et al. (2022), Bretscher,

Schmid, Sen, and Sharma (2020), and Koijen, Koulischer, Nguyen, and Yogo (2021). While

our estimation of different investor groups’ demand and the construction of instruments

follows the basic approach of this literature, we exploit the market clearing condition from

the issuer side. This identity arises naturally in our data because the supply of debt for

each country is matched to the full breakdown of holdings by investor groups. As such,

this feature contrasts with the common data structure in the literature that focuses upon

portfolio allocation of specific investors as in, for example, Maggiori, Neiman, and Schreger

(2020). We instead implement a complementary borrower-focused approach that analyzes

the investors of a given country government debt to understand the exposure of issuers.

Our paper is also related to the literature on investor demand for Advanced Economy

sovereign debt. This strand of research includes, for example, Krishnamurthy and Vissing-

Jorgensen (2012), Jiang, Krishnamurthy, and Lustig (2018), Jiang, Lustig, Van Nieuwer-

burgh, and Xiaolan (2019), Liu, Schmid, and Yaron (2020), and Liu (2022). Consistent with

the view in this literature that banks and the official sector hold government debt for liq-

uidity and other purposes, we show that holdings by these groups are relatively insensitive

to yields. We also analyze the role of the global financial intermediaries and financial con-

ditions, as highlighted by Bruno and Shin (2015), Gabaix and Maggiori (2015), Fang and

Liu (2020), and Miranda-Agrippino and Rey (2021). By contrast to these papers, we also

find that non-bank private investors increase holdings of AE sovereign debt in response to

increases in yield, thereby demonstrating the importance of non-bank investors.

5
 By studying investor types jointly, we also further the defaultable sovereign debt litera-

ture that has instead highlighted the importance of individual investor types. For instance,

an important part of the EM literature focuses only on the role of foreign investors, partic-

ularly foreign banks. See, for example, Eaton and Gersovitz (1981), Arellano (2008), Arel-

lano and Ramanarayanan (2012), Mendoza and Yue (2012), Cruces and Trebesch (2013), and

Arellano, Bai, and Mihalache (2020). On the other hand, the sovereign-bank nexus (“doom-

loop”) literature focuses on domestic bank investors as in, for example, Gennaioli, Martin,

and Rossi (2014), Perez (2014), Bocola (2016), Brunnermeier et al. (2016), Fahri and Tirole

(2018), and Chari, Dovis, and Kehoe (2020). A further area focuses on explaining the reserve

accumulation behavior of foreign official investors, as can be seen in Wooldridge (2006),

Dominguez, Hashimoto, and Ito (2012), Ghosh, Ostry, and Tsangarides (2017), Bianchi,

Hatchondo, and Martinez (2018), and Bianchi and Sosa-Padilla (2020), to name a few. We

instead evaluate the set of creditors together and thereby contribute to these literatures by

showing the relative importance of each investor group.

Lastly, we contribute to the literature by constructing a comprehensive dataset of investor

groups’ holdings to analyze the impact of investor composition on sovereign financing cost.

As noted above, this data decomposition is close to Arslanalp and Tsuda (2012, 2014), and

the extension in Arslanalp and Sunder-Plassmann (2022).3 In order to examine the impact

of marginal investors and their demand, however, we construct a dataset that begins earlier

spanning all investor types and encompasses a broader set of countries. This more expan-

sive base allows us to better estimate the impact of investor composition on sovereigns’ fi-

nancing costs. It also complements other studies based upon securities level issuances such

as Maggiori, Neiman, and Schreger (2020) and Faia, Salomao, and Veghazy (2022). Relative

to these granular data sets, our analysis provides a complete accounting of the investors of

a given sovereign and spans both domestic and foreign investors.

3 The first two of these papers also analyse a funding shock scenario, but largely emphasize the role of foreign investors.

We emphasize the role of non-bank investors, particularly noting the composition of the marginal investors (as opposed to
average holdings) as important for understanding debt financing exposure of sovereigns.

6
2 Evolving Composition of Sovereign Debt Investor Groups

We begin by analyzing the holders of sovereign debt and examining who are the marginal

investors that purchase new debt. We first describe the structure of the data and the in-

vestor group definitions, and highlight the broad trends in sovereign debt holdings. We

then analyze the marginal investment in sovereign debt for each group. We finish the sec-

tion examining non-bank investors in more depth with alternative datasets.

2.1 Investor Group Definitions

We consider three basic types of investors: (1) private banks; (2) non-banks; and (3) official

creditors. We describe briefly these three categories below before detailing their construction

in the data. We also split these investor groups by their location: foreign vs domestic. This

gives us six groups: domestic banks, foreign banks, domestic non-banks, foreign non-banks,

domestic central banks, and foreign official.

The first group is private banks. These institutions are often considered primary inter-

mediaries for debt markets including their function as primary dealers (Arnone & Ugolini,

2005). They have therefore been the focus of both the emerging market (EM) and advanced

economy (AE) branches of sovereign debt studies. In the literature on emerging market

borrowing, foreign global banks are often modeled as the primary creditor. More broadly,

liquidity and capital regulation incentivize domestic banks to hold domestic government

debt, potentially creating a bank-sovereign doom loop.4

The second group of private investors is a combination of all private investors who are

not banks. These investors are not subject to bank regulatory restrictions but may face other

constraints depending on the nature of their business. Overall, this investor group encom-

passes financial institutions such as pensions and insurance firms, endowments, mutual

funds, and hedge funds, as well as non-financial entities like corporations and households.
4 Bank regulation typically assigns risk-weights to different assets that banks hold for computing the required capital ratios.

So, acquiring sovereign debt, which typically carries a zero risk weight, does not reduce the bank’s regulatory capital ratios,
though it can affect other bank constraints like the leverage ratio. On the bank-sovereign doom loop, see Fahri and Tirole
(2018) and the Related Literature section above.

7
 Finally, we consider official creditors. “Domestic Official” creditors are simply the “Do-

mestic Central Bank”, while the “Foreign Official” group includes foreign central banks,

foreign governments, and international organizations such as the World Bank and Interna-

tional Monetary Fund.

This data set is organized around the sovereign issuers of debt, rather than tracking the

securities held by investors. Specifically, we are interested in how the holdings of each

investor group responds to changes in the sovereign’s total debt. To see this decomposition,

it will therefore be useful to define the book value of sovereign debt of a given country

indexed by n, as D (n) and the amount held by each investor group i, as Hi (n). Then, the

investor groups for country n debt can be aggregated across the investor groups to provide

a measure of the holdings of each country’s creditors as:

I
D (n) ≡ ∑ Hi (n) (1)
i =1

where I is the number of investor groups. For example, in our aggregate data set above,

I = 6 since we have three types of investors with domestic and foreign counterparts for

each. We will use the decompositions in Equation (1) to uncover the marginal ownership

responses to supply changes over time below. Before doing so, we describe the data.

2.2 Data

The annual data series for the debt and holding groups come from various sources. Here

we describe briefly the overall approach in constructing these data series, relegating a more

complete discussion to Appendix A. The general approach follows the work of Arslanalp

and Tsuda (2012, 2014), and the recent expansion of their dataset (Arslanalp & Sunder-

Plassmann, 2022). We modify their methodology in order to broaden further the time period

and sample of countries for which we have data for all investor groups. We point interested

researchers to their papers and associated database for a full description of the approach.

We first derive each sovereign’s total debt from the IMF Historical Public Debt Database

(HPDD), which provides debt-to-GDP for a large number of countries over a long time

8
horizon. We multiply this series by GDP from the World Bank to recover the value of debt in

current US dollars.5 The total foreign holdings for each sovereign are constructed following

the methodology in Avdjiev, Hardy, Şebnem Kalemli-Özcan, and Servén (2022) (hereafter,

AHKS), which combines International Investment Position (IIP) data, the Quarterly External

Debt Statistics (QEDS), and the BIS international banking and international debt securities

statistics. The domestic total holdings are computed as the difference between total debt

and foreign debt holdings.

The domestic and foreign holdings are further decomposed into the three groups de-

scribed above. Data for foreign bank holdings are estimated using the approach in AHKS.

Foreign official holdings for advanced economies and China are taken directly from Ar-

slanalp and Tsuda (2012, 2014), capturing the use of such debt as foreign reserves. For all

other countries for which their debt is less commonly held as reserves, we capture foreign

official lending as the sum of bilateral and multilateral lending from the World Bank Debtor

Reporting System (DRS). Foreign non-bank holdings are the difference between these mea-

sures for banks and foreign official with the total foreign holdings.

Sovereign debt holdings by domestic banks and domestic central banks are taken from

the IMF’s International Financial Statistics (IFS) dataset, supplemented with data from the

official websites of central banks when the data was incomplete. Domestic non-bank hold-

ings are measured as the difference between the domestic total and the sum of domestic

banks and domestic central banks. All holdings in the baseline data series are measured in

current US dollars.

Overall, this construction provides a sample of 95 countries with a breakdown of debt

by investor, with time series spanning 1991 to 2018.6 For some of the analysis, we split

the sovereigns into 3 groups: advanced economies (AEs), emerging markets (EMs), and

developing countries (DCs). Details of the country groups can be found in Appendix A.
5 For some countries, the HPDD data series stop in 2015. For these countries, we obtain post-2015 values by applying the

growth rate in total debt from the Quarterly Public Debt Statistics (QPSD), which has excellent coverage of the recent period,
to the last available level computed from the HPDD.
6 For a larger sample of 151 countries starting in 1991, partial data are available (i.e. data for some investor groups are

missing).

9
Advantages and limitations of the data. The main advantages of our dataset are that it allo-

cates all of a sovereign’s debt to individual investors, and that it spans a large number of

countries over a long time period. However, such coverage comes with drawbacks on detail.

For instance, our dataset does not include detailed information on the currency or maturity

of the debt or the specific location of the foreign investor. It does not have details on the non-

sovereign portfolios of these investors. Moreover, the data is annual, and so misses higher

frequency dynamics. In the conclusion, we discuss avenues for future research to build on

our analysis.

2.3 Investor Trends and Marginal Holders

We next examine some general trends in holdings and the relative behavior of each investor

group’s holdings as government debt changes. Panels (a), (b), and (c) of Figure 2 shows

the average investor holdings-to-GDP shares of advanced economy (AE), emerging market

(EM) and developing economies (DC) debt, respectively. These figures show distinctive dif-

ferences within the groups. For all groups, the foreign bank and non-bank shares have been

stable over time.7 However, the proportion of foreign official holdings has become larger

for AEs, as central banks have increased their holdings of safe haven government debt for

reserve purposes, while the proportion of foreign official creditors has declined for emerg-

ing markets. Strikingly, the share of domestic non-banks has increased over time for EMs

relative to AEs, perhaps reflecting growing financial development within these economies.

By contrast, the AE holdings of domestic central banks has expanded over time, tied to the

use of unconventional monetary policies and other programs.

These trends raise an important question. When the size of debt increases, which in-

vestors absorb the additional amount? In other words, who are the marginal investors for

the sovereign? To explore this question, we regress the change in debt held by each investor
7 Foreign
official data for AEs are not available until 2000, so foreign holdings are not fully reported before that point.
Coverage for EMs and DCs is broader after 1995.

10
 Figure 2: Trends in Sovereign Debt

50
80 80

40
60 60
30
40 40
20

20 20
10

0 0 0
2002 2006 2010 2014 2018 2000 2003 2006 2009 2012 2015 2018 1998 2002 2006 2010 2014 2018
DomBK DomNB DomCB ForBK ForNB ForOff

(a) Advanced economies (b) Emerging markets (c) Developing countries

Note: Aggregate debt by country group detailed in Appendix A divided by the aggregate GDP of the country group. Each
panel shows a balanced sample over the given time frame.

group on the change in total debt:

Hi,t (n) − Hi,t−1 (n) Dt ( n ) − Dt − 1 ( n )

= ai + a(n) + at + ei,t (n), ∀i (2)
Dt −1 ( n ) Dt −1 ( n )

where the subscript t indicates time, and where a(n) and at represent country fixed effects

and time fixed effects, respectively. Using the identity in Equation (1), the sum of the in-

vestor groups add up to the total so that:

I H (n) − H
Dt ( n ) − Dt − 1 ( n ) i,t−1 ( n )
= ∑ i,t . (3)
Dt − 1 ( n ) i =1
Dt −1 ( n )

Since all of the debt is absorbed by some investor, the coefficients estimated from this re-
I
gression will sum to 1: ∑ ai = 1. Each coefficient ai reflects the marginal holding response
i =1
of investor group i to variations in the supply of debt.

Table 1 shows the results of this regression. Panel A provides a baseline estimate labeled

”All” based upon a balanced sample of all countries. The first two columns provide these

results for an aggregated group of domestic and foreign investors, respectively. Columns (1)

and (2) show that for every additional unit of debt supplied, 60% is absorbed by domestic

investors while the other 40% is picked up by foreign investors. In AEs and DCs, this split

is roughly equal, whereas for EM sovereign debt, domestic investors take over two-thirds

of additional debt.

11
 Breaking down foreign and domestic investors by type in columns (3) through (8) reveals

additional insights. For the ”All” country estimates, non-bank investors tend to be the most

important, taking on 39% and 30% of additional debt holdings for domestic and foreign en-

tities, respectively. As reported in the following rows, decomposing estimates into country

groups shows the relative importance of investor groups across these countries. In partic-

ular, foreign non-banks are more important for AEs and DCs, while domestic non-banks

matter more for EMs.

Table 1: Marginal Holders of Sovereign Debt

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

Dom For DomBK DomNB DomCB ForBK ForNB ForOff
Panel A: Marginal Share

All 0.60*** 0.40*** 0.16*** 0.39*** 0.05*** 0.04*** 0.30*** 0.06***

(0.05) (0.05) (0.04) (0.05) (0.01) (0.01) (0.05) (0.02)

AE 0.50*** 0.50*** 0.02 0.39*** 0.09 0.06** 0.40*** 0.04

(0.12) (0.12) (0.03) (0.09) (0.07) (0.02) (0.12) (0.03)

EM 0.68*** 0.32*** 0.21*** 0.43*** 0.04** 0.05*** 0.24*** 0.04

(0.05) (0.05) (0.05) (0.07) (0.02) (0.02) (0.05) (0.03)

DC 0.50*** 0.50*** 0.11*** 0.34*** 0.05*** 0.01 0.40*** 0.09***

(0.09) (0.09) (0.03) (0.09) (0.02) (0.02) (0.10) (0.02)
Panel B: Average Share

All 0.57 0.43 0.22 0.28 0.07 0.06 0.18 0.19

AE 0.55 0.45 0.19 0.32 0.04 0.10 0.21 0.14

EM 0.61 0.39 0.26 0.28 0.07 0.06 0.17 0.16

DC 0.49 0.51 0.16 0.24 0.09 0.04 0.17 0.30

Note: Panel A of the table reports the regression coefficients for Equation (2) for each investor group. The first two columns
represent domestic and foreign investors, respectively. Columns (3) through (8) correspond to the six investor groups.
Standard errors clustered at the country level are reported in the parentheses. Panel B of the table reports the average share
of holding by each investor group.

More important, however, is the role of investor groups relative to each other. Here, non-

bank investors again demonstrate the most striking results. For example, foreign non-banks

play a much stronger role in expanding holdings in response to new debt than do foreign

12
banks. In particular, the take-up of new EM debt by foreign non-bank investors is 24% of

the total in contrast to only 5% by foreign banks. These findings are in stark contrast to a

view that foreign banks play the biggest role in the EM debt market. And lastly, foreign

official investors are more important as marginal investors for DCs, where they serve as an

important source of financing (9% of the increased supply). Appendix B illustrates these

patterns in panel (a) of Figure B1.

These results consider the marginal decomposition of investor groups in response to an

expansion in country debt and therefore one might wonder whether these holdings simply

represent proportional expansions to average holdings. Panel B of Table 1 reports the av-

erage holdings by investor group over the period and shows this is not the case. The large

marginal contribution of non-bank investor holdings are greater than the average holdings.

For example, in the baseline ”All” results, the average holdings of domestic non-banks and

foreign non-banks sum to only 46% (that is, 0.28 + 0.18). At the same time, the marginal

share reported in Panel A are jointly 69% (0.39 + 0.30) of the change. This pattern is clearly

robust across all the remaining decompositions including AE, EM, and DC.8

Appendix B.1 reports an extensive set of results of the same analysis under different

scenarios of recessions vs. non-recessions, crisis vs. non-crisis periods, as well as in different

subperiods. The general pattern holds in all subsamples.

The responses reported in Table 1 simply capture variations in holdings by investor

groups measured in a common currency. However, such changes in holdings may arise

from currency valuation effects for debt that is originally issued in local currency. Indeed,

the importance of valuation effects in the balance sheet adjustment of countries has been

shown in a number of papers (e.g. Gourinchas and Rey (2007)).9 To consider this possi-

bility, Appendix B.2 repeats the analysis taking into consideration the currency valuation

effect. The results are largely unchanged: non-banks have a larger response to increased
8 The most recent shares in the sample are essentially the same as the averages, which rules out the possibility that non-

banks’ average shares have an upward trend.

9 Valuation effects have also been shown to be important more generally in the portfolio allocation of investors using

disaggregated data on holdings of specific investors. See, for example, Curcuru, Thomas, Warnock, and Wongswan (2011) for
an analysis of US portfolio investment in foreign equity markets.

13
debt supply than other investor groups.

2.4 The Role of Non-Bank Investors

As we showed above, private investors who are not banks play an important role in the

sovereign debt market. This finding raises the obvious questions: who are these investors

and how do they respond to changes in sovereign debt? To shed light on this question,

we turn to other more disaggregated data sets: (i) the Euro area securities holding statistics

(SHS) produced by the European Central Bank; (ii) US TIC data; and (iii) UK Gilt Holders

data.10 We utilise the Euro area data to shed light on the foreign non-bank segment, and the

US and UK data to examine the domestic non-bank segment.

We first examine the Euro Area SHS data. The SHS provides data on the holdings of

securities for all investors located in the Euro Area, split by sector of the holder/investor,

the type (e.g., equity vs debt) of security, the sector of the issuer, and the residence of the

issuer. Thus, we can observe the holdings of government debt issued by specific countries

for all Euro area investors, disaggregated by investor type. Specifically, the non-bank sector

in this data has the more granular breakdown: (a) households and non-profits, (b) insurance

and pensions, (c) non-financial corporations; and (d) ”other financial institutions” which in-

cludes entities like hedge funds and mutual funds. As above, we also study the holdings by

banks and the official sector. However, in this data source, the official sector is the govern-

ment, excluding the central bank. These data are quarterly, with sample taken over 2013 Q4

to 2020 Q3.

This setup allows us to examine how Euro Area non-bank investors behave relative to

each other and relative to other Euro area investors in regards to their holdings of foreign

sovereign debt (i.e. outside the Euro Area). Since Euro Area investors do not account for

the full set of foreign investors for a given government, we cannot replicate our previous

analysis. Nevertheless, we can adapt that approach to establish which types of non-bank
10 Faia, Salomao, and Veghazy (2022) analyze the underlying data for the Euro Area SHS, focusing on corporate bonds in

the euro-area, and find significantly different behavior for mutual funds relative to insurance and pension funds, similar to
our results below for sovereign bonds.

14
investors in the Euro Area drive the overall behavior of Euro Area non-bank investors in for-

eign sovereign debt markets. We therefore modify our regression in Equation (2) so that the

denominator is the total holdings of foreign sovereign debt by Euro Area investors, rather

than total debt. When we examine within the non-bank group, we replace the denominator

with the total holdings of sovereign n’s debt by Euro Area non-bank investors.

The coefficient results are summarized in Figure 3 while the regression results are rel-

egated to Table B8 in Appendix B. Similar to the aggregate data, the non-bank private in-

vestors are the most marginal group within the set of Euro Area investors (excluding central

banks). For the base case of all non-Euro Area countries, 77% of every additional unit of

sovereign debt picked up by these investors is held by Non-Banks, while only 23% is held

by Banks. These general patterns hold for all cases except for Advanced Economies in the

baseline estimates shown in Figure 3 panel (a), where banks seem to absorb the largest share

of the increase.11
Figure 3: Foreign Non-bank Marginal Holders: Euro Area Investors

0.5
0.8 0.8 0.8

0.4
0.6 0.6 0.6
0.3
0.4 0.4 0.4
0.2

0.2 0.2 0.2

0.1

0.0 0.0 0.0 0.0

Marginal share Share Non-Bank marginal Marginal share Share Non-Bank marginal
Banks NonBank HH OthFin Gov I&P NFC

(a) Advanced Economies (b) Advanced Economies (c) Emerging Markets (d) Emerging Markets

Note: This figure plots the regression coefficients in Equation (2) for euro area investors’ holdings of non-euro area
government debt. Panels (a) and (c) report the coefficients for regressions of EA investor holdings of non-EU government
debt for AE governments and EM governments, respectively, reported in columns (1), (2), and (3) in Table B8. These groups
are private Banks, Non-Banks, and other governments ”Gov”. Panels (b) and (d) report the coefficients for regressions of
disaggregated Non-bank EA investor groups based upon columns (4), (5), (6), and (7) of Table B8) for Households, Insurance
and Pensions (IP), Nonfinancial Corporations (NFC), and Other Financial Institutions (OthFin). The results use the total EA
investor holdings of non-EA debt by AE and by EM separately as the total denominator in Equation (2).
11 Debt from large AE countries such as the US are important for this group of investors, but are weighted equally with

other countries. If we weight this regression by the size of the total holdings of that sovereign’s debt by European investors,
then non-banks are again the most marginal investor in foreign AE sovereign debt.

15
 More importantly, panels (b) and (d) in Figure 3 provide a decomposition within the

Non-bank investor group. Insurance and pension (IP) account for 12 to 14% of the addi-

tional sovereign debt picked up by foreign non-banks in both (non-Euro Area) advanced

and emerging markets. This is dwarfed by ”other financial institutions” (OthFin) such as

hedge funds and mutual funds, which picked up 85%. By contrast, the other two categories

of households (HH) and non-financial companies (NFC) essentially take on little to none of

the additional foreign government debt.

We next turn to the US and UK data to examine the domestic non-bank group. These

data breakout the total debt issuance by the central government by who is holding the debt.

Thus, we apply our standard approach in Equation (2), dropping the time and country fixed

effects, as we are considering a single country. These datasets provide a detailed breakdown

of domestic holders of these securities, but provide little or no sector breakdown of the

foreign holders. We utilize data over 1995 Q1 to 2020 Q4 for comparability with the other

data. While we note that the US and UK are not typical countries, so results may not be

fully generalizable, they nevertheless provide a useful window into domestic non-banks’

holdings of government debt.

The coefficients from these regressions are plotted in Figure 4 while the details are re-

ported in Tables B9 and B10 in Appendix B. For US Treasuries and UK Gilts, respectively,

Panels (a) and (c) depict the regression coefficients for the aggregated groups of total foreign

investors, domestic central banks, and domestic non-banks. As shown there, domestic non-

banks are the most marginal investors in these markets, accounting for nearly half of the

variation. Panels (b) and (d) break out the non-bank marginal share by underlying sector

for holdings of US Treasuries and UK Gilts, respectively. In the US, we see that investment

funds, including hedge funds and Money Market Funds (MMFs), are the most important

group, accounting for over half of the non-bank marginal variation (despite making up only

26% of non-bank holdings on average).12 The Insurance and Pension (I&P) sector accounts

for about 15%. In the UK, the I&P sector plays a much larger role for non-bank holdings,
12 The “household” sector includes hedge funds, private funds, and private trusts.

16
accounting for nearly half of the marginal absorption. Funds and other non-bank financial

institutions account for the other half. The large presence of I&P is notable especially in light

of the Sept 2022 turmoil in the Gilt market regarding pension funds.13

Figure 4: Domestic Non-bank Marginal Holders: US and UK

0.5
0.3 0.4 0.4

0.4
0.2 0.3 0.3
0.3
0.1 0.2 0.2
0.2

0.0 0.1 0.1

0.1

0.0 –0.1 0.0 0.0

Marginal share Share Non-bank marginal Marginal share Share Non-bank marginal
For- DomCB MMF HF/HH For- DomCB Funds I&P
eign DomBank I&P Oth Fin eign DomBank & Oth
DomNB Non Fin DomNB Non Fin

(a) US Treasuries (b) US Treasuries (c) UK Gilts (d) UK Gilts

Note: This figure plots the regression coefficients in Equation (2) for the holders of US Treasuries and the holders of UK Gilts
over 1995q1-2020q4. The regression results are reported in Tables B9 and B10, for US Treasuries and UK Gilt, respectively.
Panels (a) and (c) give the aggregate breakdown for all foreign (Foreign), domestic central bank (DomCB), domestic bank
(DomBank), and domestic non-bank (DomNB). In panel (b), “HF/HH” includes Hedge funds, private equity, private trusts,
and direct household holdings. “Non Fin” is all non-financial holders excluding households. “Oth Fin” is all other financial
institutions apart from funds and Insurance and Pension (I&P). In panel (d), “Funds & Oth” comprises all funds and
financial institutions except for I&P. “Non Fin” includes all non-financial holders.

Overall, these disaggregated data suggest that the foreign non-bank investor group is

largely driven by the behavior of investment funds, with modest participation from the

I&P sector. For the domestic non-bank investor group, the I&P sector likely plays a larger

role alongside funds. Other domestic players may be context specific (e.g., the large role of

MMFs in the US may be unique), but the results suggest that investment and other funds

are the primary non-bank players in the market.

13 The higher marginal participation of this sector is driven by their large average holdings, rather than I&P being quite

active traders. In the US, this sector makes up 43% of domestic non-bank holdings, while in the UK it accounts for 81%. This
fits with the perception that this sector is more stable in their holdings.

17
3 Sovereign Debt, Investor Holdings, and Yields

We showed above who holds sovereign debt and how the composition of investor holdings

varies when that debt expands on the margin. In this section, we describe a framework to

study the behavior of creditors, allowing us to consider why that composition matters. Sec-

tion 3.1 begins by considering how a sovereign may view the propensity for its investors to

provide credit as it varies debt and the yield on that debt. Although this framework could

in principle be used for any country, we study debt issued by AE and EM countries sepa-

rately because investors view these two types of assets differently. Investors hold AE debt

(for example, US Treasuries) mainly for its safety and liquidity, but investors primarily hold

EM debt for its yield. Therefore, in Sections 3.2 and 3.4 we estimate the funding propensi-

ties from the point of view of EM and AE sovereigns, respectively. Section 3.3 then shows

that under some additional assumptions, this framework may be used to identify and esti-

mate the investor demand similarly to the approach in Koijen and Yogo (2020). In Section

4, we combine these estimates with the marginal investor results above to determine the

sensitivity of financing cost in borrowing countries.

3.1 Sovereign Funding

We now describe a framework to relate the debt funding of a sovereign to its creditor hold-

ings and to study how those creditors respond to the yield. Clearly, the sovereign debt

market is characterized by both supply and demand, as we describe in turn.

The supply of debt depends upon factors that impact the desire to borrow along with the

cost of doing so. Specifically, a long literature has related government financing needs with

economic downturns and macroeconomic variables such as inflation (e.g. Aguiar, Chatter-

jee, Cole, and Stangebye (2020); Reinhart and Rogoff (2011)). Moreover, a government’s

ability to borrow depends upon its perceived riskiness. For example, in the defaultable

sovereign debt literature, debt levels depend upon the country’s income level and its de-

fault probability (e.g., Arellano (2008)). We summarize this relationship by treating the debt

18
supply-to-GDP as dependent on a vector of country-specific variables that impact the bor-

rowing decision, defined as xt (n) for a given country n, as well as its own lag. Thus, defining

the book value of country n’s debt as a share of GDP to be d(n) ≡ D (n)/Y (n), where Y (n)

is country n’s GDP level, we can write the debt supply function as: dt (n) = d (xt (n), dt−1 ).

Our analysis below uses the functional form for this process given by:

Dt (n) − γ0 Dt−1 (n)

≡ dt (n) − γ0 dt−1 (n) gt (n)−1 = g(xt (n)), (4)
Yt (n)

where γ0 is the legacy debt that has not matured and gt (n) = Yt (n)/Yt−1 (n) is country

n’s GDP growth rate. Governments mostly borrow long-term so that (1 − γ0 )−1 is ap-

proximately the maturity of country n’s debt. The newly issued debt depends on the set

characteristics xt (n). Below, we will assume that investors and governments view the same

country-specific variables xt (n) as exogenous to their own decisions. An interesting ques-

tion for future research is to ask whether they differ.

Given a desired supply of book value for this debt, a government then raises the debt

according to the willingness to fund by a set of creditors. To examine the funding provided

by these investors, we define Pt (n) as the price of debt of government n at time t and the

share of holdings of that country’s debt-to-GDP held by investor i at time t as:

 
Hi,t (n)
hi,t (n) ≡ d t ( n ), (5)
Dt ( n )
m ( n ) ≡ P ( n ) h ( n ). Investor i’s
Then the market value of this share can be written as: hi,t t i,t

holdings of country n’s debt are assumed to depend on the price of debt Pt (n) and other
m (n) =
country characteristics xt (n). We capture this relation as a funding function: hi,t

hi ( Pt (n), xt (n)). The market clearing condition in Equation (1) can then be rewritten as:
I
Pt (n)dt (n) = ∑ hi ( Pt (n), xt (n)). (6)
i =1

This condition implies an equilibrium price level of country n’s debt, Pt (n), as a function of

the book value of debt-to-GDP, dt (n), and the propensity for investors to fund debt depend-

ing on the characteristics of country n, xt (n).

19
 The ability for the government to sustain debt will depend upon its borrowing costs,

captured by the price of debt. In general, the price of debt depends upon the interest rate,

debt maturity and coupon repayment schedule. While this information may be available for

some individual bond issuances, the debt maturity and payment schedules for total debt are

not available. Therefore, we assume that the annualized interest cost per period is proxied

by the yield-to-maturity on pure discount government bonds. In this case, the price of debt

is: Pt (n) = (1 + yt (n))−T , where y(n) is the yield of T-year debt issued by government n.

We now use this framework to estimate the investor willingness to fund debt viewed
m ( n ) as
from the point of view of the borrowing government. We specify this function hi,t

a log-linear function of yield yt (n) and exogenous characteristics xt (n).14 The empirical

regression model of investor i’s holding shares of country n’s debt-to-income is then:

m 0
ln hi,t (n) = θ0,i yt (n) + θ1,i xt (n) + θi (n) + θi,t + ε i,t (n) (7)

As Equation (7) shows, the willingness to fund by investor group i depends not only on

the yield and country-specific characteristics x(n), but also on country and year fixed ef-

fects. Country fixed effects θi (n) are included to capture the time-invariant characteristics

of countries that affect the overall ability to fund from a given investor group.15 Year fixed

effects θi,t are included to capture worldwide economic shocks that are common to all coun-

tries. Finally, ε i,t (n) captures investor i’s latent demand for country n’s debt.

Estimating holdings shares as in Equation (7) presents two identification issues. First,

since the yield of debt is determined in equilibrium by equating demand to supply, it is

likely to be correlated with latent demand ε i,t (n). To address this problem, we construct

an instrument for yield yt (n) based on the insight of Koijen and Yogo (2020) that utilizes
14 Since Pt (n) is a monotonic transformation of yt (n), we replace Pt (n) in the Hi function with yt (n).
15 The inclusion of fixed effects is a departure from Koijen and Yogo (2019, 2020). It is motivated by the different structure
of our data: allocating all of a borrower’s debt by the investors, rather than allocating all of the investors’ assets to the
borrowers or equity issuers. Foreign investors in our data are composed of different entities for each country. The identifying
assumption of our approach is that all countries face a common willingness to fund debt by a given foreign investor group
based on their exogenous characteristics, x(n), and yield. Including a sovereign country fixed effect within the vector of
characteristics better captures constant funding differences across sovereigns for a given investor type that are not captured
by the other characteristics. Moreover, domestic investors operate in very different contexts within their own countries. Thus,
including fixed effects also help to make domestic investor behavior more comparable across borrowers by controlling for
time-invariant differences of each country within each domestic investor group.

20
the market clearing condition to construct an implied yield that is orthogonal to the latent

demand and correlated with the observed yield. Our dataset includes a full set of creditors

for a particular sovereign, which enables us to construct the instrument using this insight.

Specifically, we begin by estimating the market value of holdings per investor group as

a reduced-form function of country characteristics xt (n) using the regression:

m
ln hi,t (n) = αi0 xt (n) + αi (n) + αi,t + ui,t (n). (8)

m ( n ) = exp( α̂0 x ( n ) +
The fitted values of this regression for each investor defined as ĥi,t i t

α̂i (n) + α̂i,t ) provides an exogenous measure of the right-hand side of the market clearing

condition in Equation (6), which only depends on exogenous characteristics xt (n).

The second identification issue concerns the endogeneity of the supply of debt. There-

fore, according to Equation (4), we project current debt-to-GDP on the characteristics and

lagged debt as in the following equation:

dt (n) = γ0 dt−1 (n) gt (n)−1 + γ1 xt (n) + γ(n) + γt + et (n) (9)

Substituting the fitted values of Equation (9), defined as dˆt (n), for the realized debt along
m ( n ), into the market clear-
with the fitted values from the market value of holding shares, ĥi,t

ing in Equation (6) implies:

I
dˆt (n)(1 + ỹt (n))−T = ∑ ĥi,tm (n) (10)
i =1

where the fitted value is: dˆt (n) = γ̂0 dt−1 (n) gt (n)−1 + γ̂1 xt (n) + γ̂(n) + γ̂t . Based upon

this condition, we then solve for the hypothetical yield ỹ j,t that clears the market given the

exogenous variables. The hypothetical yield ỹ j,t is then used as an instrument for the actual

yield y j,t in the estimation of Equation (7).

3.2 Funding Emerging Market Sovereign Debt

We now use this framework above to estimate the investor funding shares for EM sovereigns.

Note that estimating Equation (7) requires specifying a set of country-specific characteristics,

xt (n). For this purpose, we draw on a large literature on emerging market sovereign debt

21
that has provided a rich set of variables that may impact investor demand. In general, credi-

tors prefer a higher yield, but are concerned about potential default risk. Thus, creditors are

more likely to invest in countries with characteristics that they view as correlated with lower

default. For example, they are likely to be attracted to countries with higher growth, but dis-

like inflation.16 In light of the literature on sovereign debt and following Koijen and Yogo

(2020) that estimates asset demand across countries, we include real GDP growth (“GDP

growth”), inflation, the logarithm of the export-to-GDP ratio (“Exp-to-GDP”), the logarithm

of the nominal GDP (”GDP”) and sovereign credit ratings (”Ratings”).17 We use the five-

year local currency government bond yield (”Bond Yield”) as the relevant price variable.18

Table 2 reports the IV estimation results, relegating the reduced form holding estimates of

Equation (8) and the debt-to-GDP estimates of Equation (9) to the Appendix Section C.1 and

C.2, respectively. As would be expected on economic grounds, all investor groups increase

holdings in response to a higher yield. Moreover, most investor groups increase demand for

a country’s sovereign debt when GDP grows and inflation declines, although insignificantly

so for some groups. Exports-to-GDP and Ratings are relatively insignificant across groups.

The responsiveness of the holdings to yield provide an important first look at the rele-

vance of different investor groups for the sovereign, as reported in the first row of Table 2.

Within domestic investors, both banks and non-banks have a significant response to yield,

but the non-bank group is almost twice as responsive. That is, in response to a 1 percent

increase in yield, domestic non-banks increase their debt holding by 23% while domestic

banks do so by 12%. Similarly, among the foreign investors, foreign non-banks are the most

responsive as they increase holdings by 38% for a 1 percent increase of yield. Thus, non-

banks appear to vary their sovereign debt holdings by more than other investor groups in
16 See for example Arellano (2008) and Aguiar and Gopinath (2007) on output growth; Arellano et al. (2020) and Reinhart
and Rogoff (2011) on inflation. Aguiar et al. (2020) provides a survey.
17 Using a data base of bilateral holdings across countries with a larger number of observations, Koijen and Yogo (2020) also

include some other variables such as imports-to-GDP, real GDP per capita, and equity volatility. Using our much smaller data
base of holdings by country, we found that these variables were generally insignificant
18 We use five-year yields because the coverage of countries is greater than that of other maturities. Similarly, we use the

yields in local currency since EM governments primarily borrow in their local currency, and have been increasingly borrowing
from abroad in local currency, thereby making the liquidity in these markets greater. For discussion, see for example Du and
Schreger (2016) and Onen, Shin, and von Peter (2023), as well as Table C4 in the BIS Debt Securities Statistics.

22
response to yield changes.

Table 2: Emerging Market Funding

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

DomBK DomNB DomCB ForBK ForNB ForCB
Bond yield 12.073∗ 22.914∗∗∗ 27.368 28.167∗∗ 38.410∗ 24.890∗
(6.833) (8.754) (17.226) (13.689) (19.704) (14.538)

GDP -0.257 0.039 -0.716 -0.403 -1.156∗ -1.692∗∗

(0.164) (0.276) (0.476) (0.329) (0.658) (0.307)

GDP growth 3.554∗ 3.949 8.568 3.325 16.068∗∗ 6.728

(2.019) (2.687) (5.413) (4.045) (6.307) (4.221)

Inflation -3.486∗∗∗ -3.471∗∗∗ -4.424∗ -6.013∗∗∗ -3.967 -4.068∗∗

(0.946) (1.304) (2.322) (1.896) (2.619) (1.976)

Exp-to-GDP -0.249 -0.266 -0.689 -0.536 -0.094 -0.362

(0.193) (0.298) (0.565) (0.387) (0.650) (0.369)

Rating 0.018 -0.001 0.049 -0.023 0.012 0.025

(0.013) (0.020) (0.035) (0.026) (0.040) (0.025)

Observations 350 323 288 350 305 342

Note: This table reports the IV estimates of the funding shares in Equation (7), with country and year
fixed effects for the EM sovereign debt. The sample spans 1996-2018 at annual frequency. The
dependent variable is holding of the group indicated in the column title to GDP. The standard errors
clustered at country and year levels are reported in the parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

These results provide insights from the sovereign’s perspective about how funding varies

across investor groups in response to changes in borrowing costs, captured by the local

currency yield. However, given that the data do not include information about the identity

of the investor, this analysis presumes governments do not have separate information about

the investor demand in the funding market. Nevertheless, in the next subsection, we show

that under some additional assumptions, our framework can be viewed as consistent with

the literature of demand system asset pricing as described in Koijen and Yogo (2019, 2020)

and our estimated demand elasticities are similar to those reported in the literature.

23
3.3 Investor Demand Estimation for Emerging Market Debt

As noted above, our data set does not provide the identity of the investor other than whether

it is domestic or foreign and its sectoral classification. Therefore, it does not provide infor-

mation about what the other holdings of these investors are. This omission is important be-

cause, in reality, investors substitute between debt issued by different countries. Moreover,

foreign investors care about returns denominated in their own currency, while domestic in-

vestors care about returns denominated in local currency. In principle, these issues could be

addressed with a more granular data set that included the identity of these investors and

their portfolios. We leave this important question for future research.

Despite these issues, under two additional assumptions, our data can be used to esti-

mate a demand system model of EM debt by different investor groups following Koijen and

Yogo (2019, 2020). We summarize these assumption below, although Appendix D provides

more details. First, we assume that all foreign investors within a specific investment group

can be treated as a single investor. This assumption requires, for instance, that a foreign

bank does not become a domestic bank within the sample. This assumption appears more

reasonable for EM sovereign debt since many of their foreign investors are from advanced

economies. Second, since we do not know the identity of the foreign investors, we assume

that these investors are all US dollar-based investors. As such, they face foreign exchange

risk in holdings of non-US dollar denominated debt and base their risk-free rate on US Trea-

suries. Although this assumption is inaccurate for some foreign investors, the continued

importance of the US dollar in the global market suggests that it provides a useful bench-

mark. We assume domestic investors are local-currency-based.

Under these assumptions, we then treat the six creditor groups as distinct investor types

with desired portfolio shares of an unobserved wealth. As in Koijen and Yogo (2020), we

model investor i’s portfolio share of country n’s debt as a logistic function of expected re-

turns and the set of country characteristics xt (n).19 That is, we denote the portfolio holdings
19 UnlikeKoijen and Yogo (2020), we do not have a breakdown of total debt holdings to wealth by investor group and
therefore we collapse the two-step decision process within and across asset classes in that paper into one single step.

24
 m ( n ) = ω ( n ) A where A is
of investor i of sovereign debt holdings of country n as: Hi,t i,t i,t i,t

the wealth of investor i in US dollars and ωi,t (n) is the share of country n’s sovereign debt in

investor i’s portfolio. Similarly, Hi,t (0) = Ai,t ωi,t (0) is the amount of investor i outside asset

holdings in US dollars and ωi,t (0) is the portfolio share of that asset. The regression model

is then given as:

m (n)
!
Hi,t
 
ωi,t (n) 0
δi,t (n) ≡ ln = ln = λ0,i µi,t (n) + λ1,i xt (n) + λi,t + λi (n) + ε i,t (n) (11)
ωi,t (0) Hi,t (0)

where λi,t and λi (n) are year and country fixed effects and µi,t (n) is the expected excess

return of country n’s sovereign debt faced by investor i. Similar to our analysis in Section

3.2, country fixed effects are included to capture time-invariant country characteristics while

time fixed effects are included to capture global-wide shocks that are common to all coun-

tries.

Estimating Equation (11) poses two challenges. First, we do not observe the net worth or

the outside asset held by investor i. Second, the expected excess return faced by domestic

and foreign investors differ according to their base currency. Therefore, we next discuss how

we approach these two challenges.

Outside Assets We first describe our approach to the outside asset for the foreign investors

and the domestic investors, in turn. As noted above, we treat foreign investors as repre-

sentative investors in the global economy who make debt allocation decisions across all the

EM countries. Given that we do not observe the wealth or outside asset holding of such in-

vestors, we assume that their outside asset holding can be absorbed by the time fixed effect

and thus not affect our parameter estimation. Clearly, this treatment requires the foreign

investor group composition to be relatively constant for all EM borrowing countries.

Domestic investors are different from foreign investors, however, as they differ by coun-

try. For example, for a particular investor group such as ”banks”, we have N distinct domes-

tic banks in this investor group, one for each country. As such, we do not have a sufficient

number of observations for a given domestic investor to identify a separate response per

25
country. Therefore, we instead treat the responses of the domestic investors jointly across

countries but condition on a proxy for the outside asset, defined for a particular domestic

investor group as Hi,t (o, n). Thus, Equation (11) can be rewritten for domestic investors as:
m (n)
!
Hi,t 0
ln = λ0,i µi,t (n) + λ1,i xt (n) + λi,t + λi (n) + ε i,t (n) (12)
Hi,t (o, n)

In our baseline results below, we consider the level of GDP as a proxy of the outside asset

held by investor i of country n, allowing comparability to the foreign investor results. Our

results are qualitatively similar based on measures such as bank credit and the equity market

cap.

Expected Excess Returns We next discuss our treatment of excess returns for foreign and

domestic investor groups. All foreign investors are assumed to be US dollar-based, as noted

earlier. Thus, for foreign investors, we follow Koijen and Yogo (2020) in treating the expected

excess return for a US dollar-based investor as the fitted value of a predictive regression of

holding period return of debt plus the exchange rate change given by:

rt+1 (n) + st+1 (n) − st (n) − yt (US) = φ f pt (n) + ψ f (st (n) − zt (n)) + χ f (n) + ν f ,t+1 (n) (13)

where rt+1 (n) is the holding return of sovereign debt of country n in local currency, st (n)

is the log of exchange rate St (n), and yt (US) is the short-term yield of US Treasuries. The

predictive variables are the log of the bond price, pt (n), and the log of the real exchange rate

given by the difference between the log nominal exchange rate, st (n), and the difference of

price levels between country n and the US, zt (n). A country fixed effect χ f (n) is included to

capture time-invariant characteristics of country n. Constructed expected excess returns are

then:

µ f ,t (n) = φ̂ f pt (n) + ψ̂ f (st (n) − zt (n)) + χ̂ f (n) (14)

By contrast, domestic investors are local-currency based and view the realized returns in

excess of their own short term rate given by predictive regressions for each currency n:

rt+1 (n) − yt (n) = φd pt (n) + χd (n) + νd,t+1 (n) (15)

26
where yt (n) is the short term rate in local currency of country n. The constructed expected

returns for domestic investors are then:

µd,t (n) = φ̂d pt (n) + χ̂d (n) (16)

Appendix D.2 provides the estimation results of the return system as well as a description

of their connection to investor demand elasticities with respect to yield.

Estimation Results Based upon these modifications for outside assets and expected returns,

we can now turn to estimating the investor demand equations. To address the endogeneity

of the expected excess return µi,t (n) for each investor group i, we first construct an instru-

ment following the same steps as in Section 3.2. These instruments are then used to estimate

the foreign and domestic investor demand in Equation (11) and Equation (12).

Table 3 provides the results of this baseline estimation in Panel A. The first two rows

again show the importance of non-bank investors, with the first row showing the depen-

dence on domestic excess returns and the second on foreign excess returns. Among the for-

eign investors, foreign non-banks are the most sensitive as a 1% change in expected excess

returns corresponds to a 10% change in debt holdings of a particular country. Among the

domestic private investors, domestic non-banks are more responsive than domestic banks

at 7.7% and 4.2%, respectively. Domestic central banks are more responsive, although this

response is likely motivated by policy decisions and not by the desire for a return. Across

investor groups, higher inflation leads to a decline in holdings while higher GDP growth

generally increases these holdings.

The estimates of expected return responses can further be used to obtain implied in-

vestor demand elasticities. Investors who increase their holdings of a particular country’s

sovereign debt in response to an increased yield would necessarily have to reduce portfo-

lio shares in another asset. This feature of the asset demand system allows for substitution

across debt issued by different countries and different assets more generally. As shown in

Koijen and Yogo (2019) and detailed in Appendix D.2, investor group i’s elasticity of de-

27
 Table 3: Investor Demand: Emerging Markets

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

DomBK DomNB DomCB ForBK ForNB ForCB

Panel A: IV Estimates

Return- Domestic 4.150∗ 7.753∗∗∗ 10.347∗

(2.190) (2.770) (6.021)

Return- Foreign 8.034∗∗ 10.056∗ 6.428∗

(3.924) (5.199) (3.844)

GDP 0.767∗∗∗ 1.099∗∗∗ 0.330 1.718∗∗∗ 1.299 0.203

(0.152) (0.252) (0.467) (0.610) (1.251) (0.587)

GDP Growth 3.739∗ 4.264 10.080∗ 5.124 17.428∗∗ 7.543

(1.968) (2.590) (5.850) (4.892) (6.982) (4.761)

Inflation -3.542∗∗∗ -3.526∗∗∗ -4.894∗∗ -7.217∗∗∗ -5.071 -4.734∗∗

(0.910) (1.235) (2.424) (2.468) (3.171) (2.398)

Exp-to-GDP -0.222 -0.216 -0.662 -0.502 -0.167 -0.304

(0.175) (0.271) (0.542) (0.400) (0.642) (0.363)

Ratings 0.016 -0.005 0.046 -0.016 0.008 0.028

(0.012) (0.018) (0.033) (0.029) (0.039) (0.026)

Panel B: Shares and Elasticities

Mean ωi (n) 0.140 0.340 NA 0.01 0.015 NA

Elasticity 10.64 14.89 NA 29.05 35.45 NA

Note: This table reports the EM sovereign debt IV estimates of the demand function in Equation (12) for the first three columns and in
Equation (11) for the last three columns, including country and year fixed effects. The sample spans 1996-2018 at annual frequency. The
dependent variable is holding of the group indicated in the column title to GDP. The standard errors clustered at country and year level are
reported in the parentheses. Panel A reports the instrumental variable estimates. * p < 0.10, ** p < 0.05, *** p < 0.01. Panel B gives the shares
and implied demand elasticities.

mand with respect to expected excess return equals λ0,i (1 − ωi,t (n)). Therefore, calculating

the elasticities requires adjusting the investors’ yield coefficients by their portfolio share of

country n debt holdings, ωi,t (n). Although we do not have measures of portfolio shares

of each investor groups, we can get estimates of domestic and foreign shares. For domes-

tic banks and domestic non-banks, we proxy for this share using the average ratio across

emerging markets of investor group holdings of own government bonds to assets. Taken

28
from the IMF Financial Soundness Indicators, these ratios are 0.14 for banks and 0.34 for

non-banks. For foreign investors, we take the data from the Euro area SHS data and BIS

banking statistics to compute proxies. In particular, holdings of EM sovereign debt by euro

area non-bank investors is about 1.5% of their portfolio. From the BIS banking statistics,

foreign banks have roughly 1% of claims on foreign sovereigns (Hardy and Zhu (2023)).

Panel B of Table 3 reports these shares and their implied elasticities with respect to yield.

These elasticities vary from 10 for domestic banks to around 35 for foreign non-banks. Al-

though much of the comparable elasticities in the literature relate to private assets instead of

sovereign debt, the estimates are consistent with other debt instruments. For example, for

ten year bonds, Jiang et al. (2022) find a range of 13 to 44, while the estimates in Koijen et al.

(2021) vary between 12 to 30 but reach 72 for foreigners. The similarity between these elas-

ticity estimates suggests that there may be a high degree of substitutability between private

and sovereign debt, an interesting question that we leave on the agenda for future research.

3.4 Funding Advanced Economy Sovereign Debt

We now return to consider the funding of Advanced Economy debt based upon the frame-

work described in Section 3.2. The approach to estimating AE sovereign debt funding is

similar to that of EM sovereigns with two modifications. First, since variations in yields are

much lower for these countries, we regress the holdings on the logarithm of the yield.20 Sec-

ond, we omit some countries. In particular, given the special status of the US Treasury Mar-

ket, we exclude the U.S. sovereign debt from the AE sample. Indeed, as is well recognized

in the literature, US Treasuries are often treated as special safe assets in the international

financial market, a status has been described as an ”exorbitant privilege.”21 Moreover, we

exclude Greece for the period of 2009 to 2015 when Greece was experiencing a debt crisis

and its yield spiked to an excessively high level.

Table 4 reports the IV estimation results for AEs. The results for the first-step regressions
20 For some countries, the yields are very low during the Zero Lower Bound period implying implausibly out-sized re-

sponses.
21 On the safe asset status, see for example, Jiang, Krishnamurthy, and Lustig (2018)and the discussion in Gourinchas, Rey,

and Govillot (2017).

29
 Table 4: Advanced Economy Funding: Instrumental Variable Estimates

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

DomBK DomNB DomCB ForBK ForNB ForCB
Bond yield 1.377∗∗∗ 2.179∗∗ 1.790∗∗ 0.979∗∗ 3.111∗∗ 0.686∗∗
(0.497) (0.865) (0.782) (0.391) (1.373) (0.301)

GDP growth 6.990∗ 4.897 15.057∗∗ 1.915 11.634 5.246∗∗

(3.726) (6.659) (6.723) (3.039) (9.600) (2.338)

GDP -0.742 -2.161∗ -1.482 0.705 0.021 0.915∗∗

(0.595) (1.130) (1.086) (0.494) (1.510) (0.380)

Exp-to-GDP 4.363∗∗∗ 5.103∗∗ 6.124∗∗ 2.957∗∗ 8.735∗∗ 1.384

(1.536) (2.408) (2.540) (1.203) (3.938) (0.926)

Rating -0.469∗ -1.128∗∗ -0.110 -0.560∗∗ -1.358∗ -0.013

(0.279) (0.498) (0.462) (0.221) (0.724) (0.170)

Inflation 3.705 4.007 20.007∗ 9.769∗∗ 8.820 11.810∗∗∗

(5.994) (9.532) (11.488) (4.940) (14.611) (3.801)

Observations 274 264 249 275 267 275

Note: This table reports the IV estimates of the funding function in Equation (7), with country and
year fixed effects. The sample spans 2000-2018 at annual frequency. The dependent variable is holding
of the group indicated in the column title to GDP. The standard errors clustered at country and year
level are reported in the parentheses.* p < 0.10, ** p < 0.05, *** p < 0.01.

for the instrument and results for the projected debt are reported in Appendices C.3 and C.4,

respectively. As the first row shows, the responses of each investor group’s holdings to yield

changes are all positive. Moreover, as with the holdings of EM debt, the non-banks are the

most responsive to changes in yields within the domestic and foreign investor groups. In

particular, a one percent change in the local currency yield leads to a 2.2% change in holding

shares by domestic non-banks compared to a 1.4% change in domestic banks. Similarly,

foreign non-banks respond to the same change with a 3.1% change in holdings compared to

a 0.98% change by foreign banks. Overall, the pattern that non-banks are more responsive

as found among emerging markets appears robust to advanced economies as well.

It would be useful to estimate these holdings as a demand system based upon the in-

vestors in the AE country government debt as we did for EM sovereigns. However, as

30
noted earlier, in our data set foreigners are only identified as living outside the given coun-

try. Thus, we cannot cleanly identify which investors are foreign for a given sovereign.

For example, German banks will be in the ”Domestic bank” group for German debt but in

the ”Foreign bank” group for French debt. Thus, we cannot estimate realistic investor de-

mand systems for AE governments with this data set. A more granular data set of portfolio

holdings of sovereign debt and other securities would be required for the estimation of the

demand system for AE sovereign debt, an important study that we leave for future research.

4 Cost of Financing Sensitivity and Counterfactual Analysis

Section 3 reported estimates of the demand response of different investors to, alternatively,

changes in the yield or the expected excess return of debt and to other country characteris-

tics. This section illustrates why the investor composition of sovereign debt holdings mat-

ters. For this purpose, we develop a measure of financing costs and show how it depends

upon the investor elasticity to yield. We then consider how this financing cost differs ac-

cording to the characteristics preferred by the various investor groups. All of the analysis in

this section is based upon the Emerging Market sample using estimates from the demand

system approach in Section 3.3. Using the estimates in Section 3.2 leads to similar results.

In this section, we conduct counterfactual analyses for private investors only. Despite

that official lenders’ holdings respond to the price of debt and other characteristics changes,

we treat official lenders’ holding as policy choices instead of market behaviors. The marginal

or average shares of official lenders are redistributed proportionally to private bank and

non-bank sectors.

4.1 Debt Increase

We begin by asking the question: how much will the cost of financing increase if the gov-

ernment wants to increase the debt by one percent? This question is clearly an important

one given the projected rise of debt-to-GDP in the next few years. To calculate this measure,

we first note that the present value of the cost of borrowing to the country is equal to the

31
present value of debt to the creditors according to the market-clearing condition in Equa-

tion (6). For expositional parsimony, we also define the market-value of debt as: Dtm ≡ Dt Pt

and subsume the dependence upon country n. Then, Appendix E.1 shows that the overall

response in the yield for a percentage increase in debt is:

dy ∑iI=1 ai /( T + ηi )
ξ≡ = (17)
d ln D m ∑iI=1 ai ηi /( T + ηi )

where ηi is the yield elasticity of demand by investor group i and ai is the change in book

value holdings by investor group i in response to changes in debt as estimated in Section

2. We call this measure ξ the ”financing cost sensitivity.” This calculation is a partial equi-

librium calculation as it does not account for substitution effects that arise from investors

adjusting their portfolios towards or away from non-EM sovereign debt assets in response

to prices.

The interpretation of this measure is straightforward. The numerator is the sensitivity of

cost of financing in response to a one percent change in the book value of debt. This sensi-

tivity equals the marginal absorption-weighted average of the inverse demand elasticities.

Intuitively, if the sovereign issues one percent more debt, ai percent is absorbed by investor

group i. To induce investor group i to absorb this amount of debt, the yield must rise by

ai /(ηi + T ). By contrast, the denominator shows how much debt should be issued at book

value in order to collect one more percent in market value. Since demand is downward-

sloping, the price of debt will in general decrease when the debt level increases. Thus, the

increase in market value of debt will be lower than the increase in book value of debt. If

the aggregate elasticity is relatively high, the price change is mild so that the increase in

debt revenue will be close to the increase in book value. We use our estimates to study the

relative importance of these effects.

Panel A of Table 5 reports these financing cost sensitivities calculated from Equation (17)

using the marginal responses a from Table 1 and elasticities η from Table 3. For convenience,

Panel D repeats the estimates used in this calculation. The first row of Panel A shows the fi-

nancing sensitivity using the total composition of investors in our sample estimates. Specif-

32
ically, if a government increases debt by 10 percent, the cost of financing given by the debt

yield, will increase by 59 basis points. The average 5-year yield of EM debt in our sample is

8.8%, so that an increase of 59 basis points represents a 6.7% increase in cost of financing.

Next, we consider how a change in the investor base would impact this financing sensi-

tivity. These counterfactuals evaluate how the financing costs are affected if the absorption

shares of a given investor group is set to zero while the remaining shares are distributed

proportionately across the remaining investors according to their marginal absorption lev-

els. As Table 5 Panel A shows in the row labeled ”No Bank”, excluding the bank investor

groups as a whole would move the cost of financing sensitivity from 59 basis points to 52

basis points in response to a 10% increase in debt revenue. By contrast, the next row la-

beled ”No Non-bank” illustrates how excluding the non-bank investor groups (domestic

and foreign) would impact financing. As reported, eliminating non-banks would signifi-

cantly increase the overall financing cost sensitivity to 80 basis points, representing a 9%

increase for the average emerging market. This greater sensitivity is due to the higher elas-

ticities of non-bank investors as reported in Panel D. Thus, when the investor base has more

non-banks, the required increase in yield for the sovereign is less.

4.2 Changes of Characteristics

We next consider how a change in the characteristics of the borrower would impact the gov-

ernment’s financing costs. Specifically, in this second set of illustrations, we ask the question:

how much would the yield have to change in response to changes in characteristics in or-

der to keep the market value of debt unchanged? This counterfactual provides information

about the investor substitutability between the borrowing costs and characteristics.

For this purpose, we define ζ as the change of yield in response to a unit change in one

of the characteristics, Xt (n), given the market value debt. Then, Appendix E.2 shows that ζ

can be expressed as:

dyt (n) ζ̃
ζ≡ = (18)
dXt (n) 1 − T ξ̃

33
where
m ( n )) m (n)
! !
I ∂ ln Hi,t I ∂ ln Hi,t
1
ζ̃ = −
T ∑ ∂Xt (n)
ψi ÷ 1− ∑
∂Pt (n)
ψi
i =1 i =1
and where ψi is the average share of emerging market debt held by investor group i, as given

in Table 1. The numerator ζ̃ is the response of yield to a unit change of characteristic Xt (n)

for a given book value of debt Dt (n). However, if the yield of debt changes in response to

changes in X, the market value of debt will move accordingly. Therefore, the book value of

debt should adjust to offset the yield change as captured by dividing by (1 − Tξ ).

To analyze this relationships, we focus on two characteristics: real GDP growth and the

inflation rate. Panel B of Table 5 reports the response of yield to a one percent decrease

in real GDP growth. As shown in Table 3 (repeated in Table 5 Panel D), a decrease of real

GDP growth induces investors to reduce their demand of debt, so that the yield increases.

In response to a one percentage point decrease of real GDP growth, the yield increases by

36 basis points, which is equal to a 4.1% rise in the average yield to EM debt. Similarly, if

inflation increases by one percent, the yield would have to increase by 22 basis points, which

is equal to 2.5% of the average yield to EM debt. The reason is that, other things equal, all

investors dislike investing in the debt of countries with low GDP growth and high inflation.

To evaluate how investors differ according to their view of these attributes and their im-

pact on funding, we next consider how the the cost of financing sensitivity changes with

different investor bases. Table 5 Panel B shows the results of this analysis under the rows

labeled ”No Bank” and ”No Non-bank.”22 Excluding the bank investor groups as a whole,

yield will increase by 41 basis points in response to a one-percent real GDP growth decrease

and 19 basis points in response to a one-percent inflation increase. By contrast, excluding the

non-bank investor groups implies that the yield would have to increase by 29 basis points

to a one percent decrease in real GDP growth but a significantly higher 28 basis points to an

increase in inflation. As Panel D shows, the reason is that foreign non-banks’ holdings of

debt have the strongest response to GDP growth and foreign banks’ holdings of debt have
22 As before, the absorption shares of different investor groups are set to zero with the remaining shares distributed propor-
tionately across the remaining investors according to their marginal absorption.

34
the strongest response to inflation among all investors. Thus, excluding investor groups

with strongest responses reduces the yield pressure from these changes. Overall, these re-

sults highlight the importance of non-banks not only for their sensitivity to yield, but to

characteristics as well.

4.3 Changes of Investors’ Latent Demand

Finally, we analyze the impact of changes in investor group latent demand on the borrow-

ing costs. In particular, we ask: how much would yield change in response to a one percent

increase in a given investor group’s latent demand given the market value of debt? Intu-

itively, if investors in a particular group, k, increase their latent demand, the price of debt

will increase. Thus, the government can issue less book value of debt to maintain the same

market value. The reduced debt issuance will further lower debt yield. Appendix E.3 shows

that this sensitivity of financing costs to latent demand can be expressed as:

dyt (n) ϕ̃
ϕ≡ = (19)
dek,t (n) 1 − Tξ
 
where ϕ̃ = ψk /T ÷ 1 + 1
T ∑iI=1 ψi ηi . Table 5 Panel C reports the results of this analysis.

Since domestic investors are in general less elastic than foreign investors and also hold more

debt, a demand increase leads to the larger yield reduction for domestic investors than for-

eign investors. An increase in demand by foreign non-banks generates a sizable shift in

costs despite holding a lower share.

5 Concluding Remarks

The rising levels of government debt worldwide in the wake of the Covid-crisis have made

urgent the answers to questions about their repayment. At the front of those questions is:

who holds this debt and does it matter to borrowers in this market? In this paper, we address

these questions by analyzing a unique data set that decomposes sovereign debt into investor

holding groups for a large number of countries over almost three decades.

Based upon our analysis, the answers to these questions are striking. First, private fi-

35
 Table 5: Borrowing Cost Sensitivity and Counterfactuals

Panel A: Borrowing Cost Sensitivity to Debt Increase

Investor Composition Sensitivity a % Yield change b

Actual 0.59 6.70%

No Bank 0.52 5.91%

No Non-bank 0.80 9.09%

Panel B: Borrowing Cost Sensitivity to Characteristics

Investor Composition Real GDP growth Inflation

Sensitivity % Yield change b Sensitivity % Yield change b

Actual 0.36 4.09% 0.22 2.50%

No Bank 0.41 4.66% 0.19 2.16%

No Non-bank 0.29 3.30% 0.28 3.18%

Panel C: Borrowing Cost Sensitivity to Latent Demand

Investor Group Domestic Bank Domestic Non-bank Foreign Bank Foreign Non-bank

Sensitivity -1.32 -1.42 -0.30 -0.86

Panel D: Components from Estimates

Component Domestic Bank Domestic Non-bank Foreign Bank Foreign Non-bank

Elasticity η 10.64 14.89 29.05 35.45

Marginal Shares a 0.21 0.43 0.05 0.24

Average Shares ψ 0.26 0.28 0.06 0.17

GDP growth: ∂H/∂X 3.74 4.26 5.12 17.43

Inflation: ∂H/∂X -3.54 -3.53 -7.22 -5.07

Panel A reports calculations of the borrowing cost sensitivity in Equation (17). Panel B gives the effects on borrowing costs
due to the characteristics as in Equation (18). Panel C provides the effects of latent demand by investor group given in
Equation (18). Panel D provides the measures used in the counterfactual calculations. The marginal shares a and average
shares ψ are from Table 1 while the semi-elasticities η, and coefficients for GDP growth and inflation are from Table 3.
Footnotes: a scaled by 10% ; b based upon average EM five-year yield in our sample of 8.8%

36
nancial institutions that are not banks absorb substantially more of the variation in out-

standing government debt than other investor groups. Further decomposing this non-bank

investment group using country/region specific data, we find that investment funds are

the primary drivers of this larger group. Next, we identify the propensity to provide fund-

ing across investor groups by exploiting the market clearing condition provided by the full

set of creditors in our data. Using these estimates, we find that the elasticity of demand

by non-bank investors is higher than other groups. Finally, we use both the marginal in-

crease of holdings together with EM investor demand estimates to calculate the financing

cost sensitivities. An average EM country faces significant financing cost sensitivity since a

10% increase would lead to a nearly 7% increase in borrowing costs. These countries face

the greatest sensitivity of financing cost against losing non-bank investors compared to any

other investment groups, as removing them from the investor pool increases the borrowing

cost response from 7% to 9%. We conclude that EM sovereign investors are highly vul-

nerable to the presence or absence of non-bank investors. Thus, the behavior of non-bank

investors is crucial for understanding sovereign debt sustainability.

Our analysis opens up several avenues for future research, particularly given the limita-

tions of our data. For example, our data is annual and lacks detailed data on the currency

and maturity structure of sovereign debt. It also does not report the location of foreign

investors, identities of investors, or the non-sovereign portion of investor portfolios. Data

built from the investor side, such as fund level data or granular data underlying the euro

area SHS data, would allow researchers to dig further into the aspects of the data missing

from our study and provide greater analysis from the investors’ perspectives. They would

allow a deeper understanding of how investors trade off their holdings of sovereign debt

with other assets, how they respond differently to local currency versus foreign currency

debt of emerging market sovereigns, how the yield elasticity varies by maturity, or how in-

terventions by AE central banks affect their presence in the market. They would also allow

for greater analysis of the heterogeneity of this very diverse group of investors. We leave

these important questions on the agenda for future research.

37
References

Aguiar, M., Chatterjee, S., Cole, H., & Stangebye, Z. (2020). Self-fulfilling debt crises, revisited. Federal

Reserve Bank of Philadelphia Working Paper, No 20-03.

Aguiar, M., & Gopinath, G. (2007). Emerging market business cycles: The cycle is the trend. Journal of

Political Economy, 115(1), 69-102.

Arellano, C. (2008). Default risk and income fluctuations in emerging economies. American Economic

Review.

Arellano, C., Bai, Y., & Mihalache, G. P. (2020). Monetary policy and sovereign risk in emerging economies

(nk-default) (Tech. Rep.). National Bureau of Economic Research.

Arellano, C., & Ramanarayanan, A. (2012). Default and the maturity structure in sovereign bonds.

Journal of Political Economy, 120(2), 187-232.

Arnone, M., & Ugolini, P. (2005). Primary dealers in government securities. International Monetary Fund,

Washington, D.C..

Arslanalp, S., & Sunder-Plassmann, L. (2022). Sovereign debt repatriation during crises. IMF Working

Paper, No. 2022/077.

Arslanalp, S., & Tsuda, T. (2012). Tracking global demand for advanced economy sovereign debt. IMF

Working Paper, No. 12/284.

Arslanalp, S., & Tsuda, T. (2014). Tracking global demand for emerging market sovereign debt. IMF

Working Paper, No. 14/39.

Avdjiev, S., Hardy, B., Şebnem Kalemli-Özcan, & Servén, L. (2022). Gross capital flows by banks,

corporates, and sovereigns. Journal of the European Economic Association, 20(5).

Bianchi, J., Hatchondo, J. C., & Martinez, L. (2018). International reserves and rollover risk. American

Economic Review, 108(9), 2629–2670.

Bianchi, J., & Sosa-Padilla, C. (2020). Reserve accumulation, macroeconomic stabilization, and sovereign risk

(Tech. Rep.). National Bureau of Economic Research.

Bocola, L. (2016). The pass-through of sovereign risk. Journal of Political Economy, 124(4), 879–926.

Bretscher, L., Schmid, L., Sen, I., & Sharma, V. (2020). Institutional corporate bond pricing. Swiss Finance

Institute Research Paper(21-07).

Brunnermeier, M., Garicano, L., Lane, P., Pagano, M., Reis, R., Santos, T., . . . Vayanos, D. (2016). The

sovereign-bank diabolic loop and esbies. American Economic Review, 106(5), 508-512.

Bruno, V., & Shin, H. S. (2015). Cross-border banking and global liquidity. Review of Economic Studies,

38
 82(2), 535-564.

Chari, V. V., Dovis, A., & Kehoe, P. J. (2020). On the optimality of financial repression. Journal of Political

Economy, 128(2), 710–739.

Cruces, J., & Trebesch, C. (2013). Sovereign defaults: the price of haircuts. American Economic Journal:

Macroeconomics, 5(3), 85-117.

Curcuru, S. E., Thomas, C. P., Warnock, F. E., & Wongswan, J. (2011). Us international equity investment

and past and prospective returns. American Economic Review, 101(3440-55).

Dominguez, K., Hashimoto, Y., & Ito, T. (2012). International reserves and the global financial crisis.

Journal of International Economics, 88(2), 388-406.

Du, W., & Schreger, J. (2016). Local currency sovereign risk. Journal of Finance, 71(3), 1027-1069.

Eaton, J., & Gersovitz, M. (1981). Debt with potential repudiation: Theoretical and empirical analysis.

Review of Economic Studies, 48(2), 289–309.

Fahri, E., & Tirole, J. (2018). Deadly embrace: Sovereign and financial balance sheets doom loops.

Review of Economic Studies, 85(3), 1781–1823.

Faia, E., Salomao, J., & Veghazy, A. V. (2022). Granular investors and international bond prices: Scarcity-

induced safety. Available at SSRN.

Fang, X., & Liu, Y. (2020). Volatility, intermediaries, and exchange rates. Journal of Financial Economics.

Gabaix, X., & Maggiori, M. (2015). International liquidity and exchange rate dynamics. Quarterly Journal

of Economics, 1369-1420.

Garcia Luna, P., & Hardy, B. (2019). Non-bank counterparties in international banking. BIS Quarterly

Review, September, 15-31.

Gennaioli, N., Martin, A., & Rossi, S. (2014). Sovereign default, domestic banks, and financial institu-

tions. Journal of Finance, 69(2), 819-866.

Ghosh, A., Ostry, J., & Tsangarides, C. (2017). Shifting motives: Explaining the buildup in official

reserves in emerging markets since the 1980s. IMF Economic Review, 65, 308–364.

Gourinchas, P.-O., & Rey, H. (2007). International financial adjustment. Journal of Political Economy,

115(4), 665-703.

Gourinchas, P.-O., Rey, H., & Govillot, N. (2017). Exorbitant privilege and exorbitant duty. Working

Paper.

Hardy, B., & Zhu, S. (2023). Covid, central banks and the bank-sovereign nexus. BIS Quarterly Review,

March, 15-31.

39
Jiang, Z., Krishnamurthy, A., & Lustig, H. (2018). Foreign safe asset demand for us treasurys and the

dollar. AEA Papers and Proceedings, 108, 537-541.

Jiang, Z., Lustig, H., Van Nieuwerburgh, S., & Xiaolan, M. Z. (2019). The u.s. public debt valuation

puzzle. Working Paper.

Jiang, Z., Richmond, R., & Zhang, T. (2022). A portfolio approach to global imbalances. Journal of

Finance, Forthcoming.

Koijen, R. S., Koulischer, F., Nguyen, B., & Yogo, M. (2021). Inspecting the mechanism of quantitative

easing in the euro area. Journal of Financial Economics, 140(1), 1-20.

Koijen, R. S., Richmond, R. J., & Yogo, M. (2020). Which investors matter for equity valuations and

expected returns? Working Paper.

Koijen, R. S., & Yogo, M. (2019). A demand system approach to asset pricing. Journal of Political Economy,

127(4), 1475–1515.

Koijen, R. S., & Yogo, M. (2020). Exchange rates and asset prices in a global demand system (Tech. Rep.).

National Bureau of Economic Research.

Krishnamurthy, A., & Vissing-Jorgensen, A. (2012). The aggregate demand for treasury debt. Journal of

Political Economy, 120(2).

Laeven, L., & Valencia, F. (2020). Systemic banking crises database ii. IMF Economic Review, 68(2),

307–361.

Liu, Y. (2022). Government debt and risk premia. Journal of Monetary Economics, Forthcoming.

Liu, Y., Schmid, L., & Yaron, A. (2020). The risks of safe assets. Available at SSRN 3699618.

Maggiori, M., Neiman, B., & Schreger, J. (2020). International currencies and capital allocation. Journal

of Political Economy, 128(6), 2019-2066.

Mendoza, E., & Yue, V. (2012). A general equilibrium model of sovereign default and business cycles.

Quarterly Journal of Economics, 127(2), 889-946.

Miranda-Agrippino, S., & Rey, H. (2021). Us monetary policy and the global financial cycle. Review of

Economic Studies.

Onen, M., Shin, H., & von Peter, G. (2023). Overcoming original sin: insights from a new dataset. BIS

Working Papers, No 1075.

Perez, D. J. (2014). Sovereign debt, domestic banks and the provision of public liquidity.

Reinhart, C. M., & Rogoff, K. S. (2011). This time is different: Eight centuries of financial folly. In This

time is different: Eight centuries of financial folly.

40
Wooldridge, P. (2006). The changing composition of official reserves. BIS Quarterly Review, September

2006, 25-38.

41
 Online Appendix

A Data construction

A.1 Debt and Investor Holdings

The series in the aggregate dataset are sampled at an annual frequency, covering the years between

1990 and 2018. The series are denominated in US dollars.1 Debt for each category refers to general

government debt, which consists of state, local, and central government debt.

Total Total debt holdings are measured by combining the data from the IMF Historical Public Debt

Database (debt-to-GDP) and GDP series from The World Bank. While a full data series over time are

available for some, there are 96 countries for which the debt-to-GDP series ends in 2015.2 For these

countries, we forecast the total debt level using the forward-looking growth rates from QPSD total debt

series in years 2016-2018.

Foreign Total The methodology for calculating foreign total holdings is based on that in Avdjiev et

al. (2022). We start by obtaining outstanding external liabilities of the general government sector from

the International Investment Position (IIP) data. This consists of adding together liability positions

from both portfolio debt securities (bonds) and other investment debt (loans). However, coverage of

IIP data by sector can be sparse, especially for recent years (Avdjiev et al., 2022). So, we fill in missing

values from a number of sources. First, if only data for the government sector is missing, but other

sectors are reported, we fill the missing government sector internally from the IIP by subtracting the

reporting sectors from the total. Then, the Quarterly External Debt Statistics (QEDS), when available

(for a limited set of countries starting from 2005). To fill in remaining values, we utilize data from the

Bank for International Settlements (BIS).

1 Series
that are originally reported in local currency are converted to US dollars using the end of period exchange rate
from the IMF. Robustness to currency value effects are described in Section B.2
2 Afghanistan, Algeria, Angola, Anguilla, Antigua and Barbuda, Argentina, Armenia, Aruba, Azerbaijan, Bahamas,

Bahrain, Belize, Benin, Bermuda, Bhutan, Bolivia, Botswana, Brunei Darussalam, Burkina Faso, Burundi, Cabo Verde,
Cameroon, Cayman Islands, Central African Republic, Chad, Comoros, Republic of Congo, Cote d’Ivoire, Curacao, Djibouti,
Dominica, Ecuador, Republic of Equatorial Guinea, Eritrea, Eswatini, Gabon, Gambia, Ghana, Grenada, Guinea, Guinea-
Bissau, Guyana, Haiti, Hong Kong SAR, Iran, Iraq, Jamaica, Jordan, Kenya, Kuwait, Lao People’s Democratic Republic,
Lesotho, Liberia, Libya, Liechtenstein, Macao SAR, Madagascar, Malawi, Maldives, Mali, Mauritania, Morocco, Mozambique,
Myanmar, Namibia, Nepal, New Zealand, Niger, Oman, Pakistan, Papua New Guinea, Qatar, Rwanda, Samoa, San Marino,
São Tomé and Prı́ncipe, Saudi Arabia, Senegal, Sierra Leone, Singapore, Solomon Islands, South Africa, Sri Lanka, Sudan,
Suriname, Syrian Arab Rep., Togo, Tonga, Trinidad and Tobago, Tunisia, Uganda, Uruguay, Vanuatu, Venezuela, Zambia,
Zimbabwe.

i
 To fill in missing values for outstanding amounts of externally held government bonds, we use the

BIS International Debt Securities (IDS) data. The IDS consists of all debt securities issued in interna-

tional markets, which target foreign investors.3 These data are available for a wide range of countries

and time periods, and include granular splits by issuer country and sector, as well as other details not

utilized in this paper. Internationally issued bonds serve as a good proxy for bonds held by foreign

investors, especially in the case of emerging markets.4

To fill in missing values for outstanding amounts of non-bond external sovereign liabilities, we use

the BIS International Banking Statistics, to capture lending from foreign banks, as well as IMF credit

from the IIP, which captures official lending. The foreign bank estimation is described below.

Foreign Official Foreign official holdings for advanced economies and China are taken from Ar-

slanalp and Tsuda (2012) and Arslanalp and Tsuda (2014), consisting mostly of foreign official reserves

held abroad. The remaining countries are populated with the data from the World Bank debtor re-

porting system (DRS) data on bilateral and multilateral official lending to emerging and developing

economy governments.

Foreign Bank The methodology for estimating foreign bank holdings is based on Avdjiev et al. (2022).

It leverages two datasets on international banking activity compiled by the BIS.

The Locational Banking Statistics (LBS) capture outstanding claims and liabilities of internationally

active banks located in 44 reporting countries against counterparties residing in more than 200 coun-

tries. Banks record their positions on an unconsolidated basis, including intragroup positions between

offices of the same banking group. The data are compiled based on the residency principle (as done for

IIP or QEDS). The LBS capture the overwhelming majority of cross-border banking activity. The his-

torical LBS data break down counterparties in each country into banks (bank and central bank sectors)

and non-banks (corporate and government sectors).5

The second set of banking data is the Consolidated Banking Statistics (CBS). This differs from the

LBS in that the positions of banks reporting to the BIS are aggregated by the nationality (rather than by
3 A bond is flagged as international if the registration of the bond, the governing law of the bond, or the listing location

of the bond are not the same as the bond issuer’s country of residence. This classifies in essence any bond whose issuance
targets investors outside of its local market.
4 For US government bonds, when missing from IIP or QEDS, we utilize TIC data to fill in missing values, since US gov-

ernment bonds are issued domestically and frequently bought by foreigners.

5 Data on total cross border claims by BIS reporting banks separated by bank and non-bank counterparties are available

going back to 1978. More recent enhancements to the BIS LBS data (since 2013) provide more granular counterparty sector
splits, dividing the non-bank sector into the non-bank private sector and the public sector (ie government) (Garcia Luna &
Hardy, 2019).

ii
the residence) of the reporting bank.6 We use the CBS on an immediate counterparty basis (CBS/IC).7

The CBS data historically provide a borrower breakdown of the Non-Bank Sector into Public and Pri-

vate.

While we maintain a residence perspective when identifying holders of government debt, consis-

tent with the LBS and IIP, the CBS have a longer time series of the breakdown between public and

private borrowers, which allow for a larger and longer sample of estimates for foreign banks’ lending

to governments. We use the share of international bank debt for each sector from the CBS to estimate

the split of the Non-Bank LBS data into Public and Private components.8 We calculate the public sector

borrowing from cross-border banks as follows:

I NTC pub,n,t
XBS pub,n,t = XBCnb,n,t
[ (A.1)
I NTCnbp,n,t + I NTC pub,n,t

where npb indicates Non-Bank Private, nb indicates Non-Bank, pub indicates Public, n denotes the

borrowing country, and t denotes the time period. [

XBS is our estimated cross border bank holdings

of debt, XBC denotes the cross border claims (from the LBS) of BIS reporting banks, and I NTC is

international claims (from the CBS on immediate counterparty basis). The CBS international claims are

defined as the sum of XBC and the local claims by foreign affiliates of these banks that are denominated

in foreign currencies (LCFC).The sector breakdown in the CBS is not available for cross-border claims,

only for international claims.

This construction of the split of bank debt makes the following assumptions: First, the sectoral

shares for I NTC are the same as the sectoral shares for XBC. This is reasonable since for most countries,

LCFC tends to be small relative to XBC.9 Second, the sectoral shares for the set of banks that report LBS

data (44 countries) are the same as the sectoral shares for the set of banks that report CBS data (31

countries). The 31 CBS reporting countries account for about 90% of the XBC in the LBS, and the CBS

captures the activities of the subsidiaries of banks from these 31 countries worldwide. As a result, the

CBS data are sufficiently representative to make the above assumption a reasonable one. Third, data
6 For example, the positions of a French bank’s subsidiary located in New York - which in the LBS are included in the
positions of banks in the United States - are consolidated in the CBS with those of its parent and are included in the positions
of French banks.
7 The CBS are compiled in two different ways: by immediate counterparty and by ultimate risk. The immediate counter-

party is the entity with whom the bank contracts to lend or borrow. Ultimate risk takes account of credit risk mitigants, such
as collateral, guarantees and credit protection bought, which transfer the bank’s credit exposure from one counterparty to
another.
8 This general approach is also used in Arslanalp and Tsuda (2012) and Arslanalp and Tsuda (2014)
9 While for most countries, LCFC tends to be small relative to XBC, there are a small number of exceptions. For example,

this is not the case in dollarized economies (e.g. Ecuador) and some emerging European economies (e.g. Hungary and
Poland), where lending denominated in euro and in Swiss francs has been non-negligible.

iii
for the CBS that allow us to estimate the split of Non-Bank into Public and Private are not available for

advanced economies before 2000, and are only available on a semiannual basis for EM for the period

before 2000. We linearly extrapolate the semiannual shares to Public and Private into a quarterly series

for EM. For advanced economies, we assume constant shares from 2000 backwards.10

Recently, the BIS has released its enhanced banking data, starting in 2013. These data contain more

granular borrowing sector splits - Bank, Public, and Non-Bank Private (Garcia Luna & Hardy, 2019).

Avdjiev et al. (2022) use this short, recent series to establish that this methodology for estimating bor-

rowing sector splits generates estimates that are very close to the actual (reported) figures.11 In addition,

we make a correction for Switzerland where holdings by external banks are significantly overestimated

with our methodology.12

Foreign Non-bank The Foreign Non-bank series is computed by subtracting Foreign Official and

Foreign Bank series from the Foreign Total.

Domestic Total Domestic Total series is computed by subtracting Foreign Total from the Total.

Domestic Central Bank For the most part, domestic central bank holdings are taken from the IMF’s

International Financial Statistics (IFS) data set. This data base provides the debt holdings levels from

the Standardized Reporting Form (SRF) only from 2001 onwards. Therefore, debt holding levels prior

to 2001 are backcasted with annual growth rate taken from the non-standardized reporting form (non-

SRF) in the same dataset. For the countries where the IFS data were incomplete, additional data were

taken from the official websites of respective central banks.13 For these cases, the IFS data were supple-

mented using the backward-looking growth rates taken from central banks’ websites.

Domestic Bank These holdings were compiled using the same procedure as for the Domestic Central

Bank.14
10 The assumption of constant shares for advanced economies before 2000 is not too concerning, as we only extend back 4
years.
11 Since not all LBS reporting countries have started providing the enhanced borrowing sector splits, these comparisons are

based on the set of LBS reporting countries which had started reporting enhanced LBS data as of March 2016.
12 Specifically, the ratio from consolidated banking statistics (CBS) is close to 30% around 2014, while updated data from the

Locational Banking Statistics (LBS), which includes a sector breakdown for government lending in recent years, suggests the
true ratio is closer to 10%, but not more than 20%. We therefore use a ratio of 15% to compute foreign bank holdings of Swiss
government debt.
13 Austria, Belgium, Bulgaria, Finland, France, Germany, Greece, Iceland, Ireland, Korea, Latvia, Portugal, Spain, Sweden,

UK.
14 The list of countries for which additional data from the official Central Bank websites was used: Belgium, Finland, France,

Germany, Korea, Luxembourg, Mexico, Netherlands, Portugal, Serbia, Spain, Sweden.

iv
Domestic Non-bank The Domestic Non-bank series were computed by subtracting the Domestic

Central Bank and Domestic Bank series from the Domestic Total.

Inconsistencies and Cleaning When combining data across different sources, inconsistencies are in-

evitable. While most of the dataset fits together, there are some cases where the sum of some of the

components (e.g. domestic central bank and domestic bank) add to more than the total (e.g. domes-

tic total). In these cases, the procedure produces some negative observations for residually computed

groups (e.g. domestic non-banks). In general, we used the following procedures to maintain internal

consistency in the dataset (i.e. the sum of the parts add up to the whole) for these special cases.

If the Foreign Official plus Foreign Bank is greater than the Foreign Total, we replaced the Foreign

Total as the sum of the Foreign Bank plus the Foreign Official; that is, replace Foreign = max(Foreign

total, foreign official + foreign bank).

If the sum of the Foreign total and the Domestic bank and the Domestic Central Bank is greater than

the total debt, we replace total debt as this sum; that is, replace Total debt = max(Total debt, foreign

total + domestic bank + domestic Central Bank)

Given these updated variables, we compute any residual categories as needed; that is, we subtract

the other variables from the updated totals to measure the Foreign Non-bank, the Domestic Total, and

the Domestic Non-bank.

After following this process, all of the generated data series are greater than or equal to zero, and

the data are internally consistent. Further, we manually examine cases where the negative values were

large to make sure that this procedure made sense. In a few cases where it appears driven by low data

quality, we replace the observation with linear interpolation.

A.2 Country Groups

The final sample consists of the following 95 countries:

Advanced economies (23): Australia, Austria, Belgium, Canada, Denmark, Finland, France, Ger-

many, Greece, Iceland, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Portugal, Singapore,

Spain, Sweden, Switzerland, United Kingdom, and the United States.

Emerging markets (45): Argentina, Bahrain, Bangladesh, Belarus, Brazil, Bulgaria, Chile, China,

Colombia, Croatia, Cyprus, Czech Republic, Estonia, Egypt, Hungary, India, Indonesia, Israel, Jordan,

Latvia, Lithuania, Kuwait, Malaysia, Mexico, Morocco, Nigeria, Oman, Pakistan, Peru, Philippines,

v
Poland, Qatar, Romania, Russia, Serbia, Slovak Republic, Slovenia, South Africa, South Korea, Thai-

land, Turkey, Uruguay, Ukraine, Venezuela, and Vietnam.

Developing economies (27): Jamaica, Paraguay, Sudan, Angola, St. Lucia, Lebanon, Bolivia, Tunisia,

Papua New Guinea, Panama, Kenya, Sri Lanka, El Salvador, Ecuador, Dominican Republic, Malta, Al-

bania, Montenegro, Costa Rica, Namibia, Gabon, Kazakhstan, Liberia, Mongolia, Guatemala, Bahamas,

and Cote d’Ivoire.

A.3 Financial, Price, and Characteristics Data

We now describe the data sources for the financial data used in Section 3. The 5-year bond yields are

the corresponding zero coupon yields provided by the Global Financial Data. Countries with avail-

able yields are Korea, United Kingdom, Venezuela, Sri Lanka, Iceland, Norway, Sweden, Australia,

Germany, Singapore, Denmark, Japan, United States, France, Belgium, New Zealand, Canada, Spain,

Finland, Hong Kong, Hungary, Portugal, Malaysia, Switzerland, Italy, India, Morocco, Philippines,

Netherlands, Thailand, Ireland, Austria, Greece, South Africa, Mexico, Poland, Mauritius, Armenia,

Fiji, Pakistan, Tunisia, Zambia, Belarus, Bulgaria, Bangladesh, Malta, Brazil, China, Indonesia, Viet-

nam, Angola, Colombia, Croatia, Israel, Peru, Qatar, Romania, Russia, Slovenia, Turkey, Egypt, Lithua-

nia, Latvia, Uganda.

For relevant short term rates by currency, we try to obtain 3-month government bond yields where

possible and, if unavailable, we use instead short term deposit rates in local currency. In particular, we

implement the following procedure. First, we take the 3-month bond yields if available from Global

Financial Data. Next, if missing, we use instead the short-term (most of them are 3-month) deposit

rate from the Global Financial Data. Then, if short term rates are still missing, we use the deposit

rate available from the World Bank (World Development Indicators, FR.INR.DPST). Finally, for the few

countries where no short rates are available from these other sources, we fill in the missing data with

the 1-year yield downloadable from the Global Financial Data.

Exchange rates are obtained from the World Bank (Global Economic Monitor, DPANUSSPF). Prices

used for the real exchange rates are downloaded from World Bank. We use the 4th quarter Headline

Consumer Price Index as the inflation of each year.

The characteristics data used are described next. We obtained real GDP growth, inflation, exports,

and the GDP level from the IMF World Economic Outlook. In particular, GDP growth is from se-

ries NGDPR PCH, the inflation is series PCPIEPCH, and the GDP level is from series NGDPD. The

vi
Sovereign Ratings measure is a combination of SP Sovereign Rating and Fitch Sovereign Rating. In or-

der to convert the discrete series to numerical levels, we follow the procedure described in Koijen and

Yogo (2020).

B Marginal Investor Analysis: Additional Results

This appendix provides more detailed results for the analysis in Section 2.

B.1 Marginal investors during different circumstances

In this appendix subsection, we report robustness checks for the analysis in Section 2.3. For example,

the analysis in the text treats the investor composition as constant. However, in reality, the composition

of marginal investors may change depending on the time period or circumstance. Indeed, the literature

on marginal investors has highlighted their importance and differences during crises and recessions

(e.g., Bruno and Shin (2015) and Miranda-Agrippino and Rey (2021).) During these times, banks may

cut back lending and central banks may intervene to stabilize the economy. Moreover, these cut-backs

may impact the overall responses of investor holdings of government debt. Therefore, we examine the

sensitivity of investor group holdings to crises and other special circumstances.

To examine marginal investor responses during these time periods, we estimate Equation (2) sepa-

rately when the country-year is, alternatively, (i) in a recession and not; (ii) during a banking crisis and

not; and (iii) in different sub-periods.15 We consolidate the results of these regressions into Figures B1

(panels (b) and (c)) and Figures B2, relegating detailed reporting of the regression coefficients to Tables

B1, B2, and B3.

The figures highlight important patterns across geographic groups. Both Advanced Economies and

Emerging Market sovereign debt show marked differences in their marginal investors across different

circumstances. For example, domestic non-banks absorb more emerging market debt during recessions

or banking crises, whereas domestic banks decrease their absorption. In advanced economies, the do-

mestic central bank becomes a key investor during a crisis. After the Global Financial Crisis (GFC), both

domestic banks and domestic non-banks increase the share of debt they absorb. A similar pattern holds

for foreign official lenders in the case of advanced economies. This latter trend may reflect increased
15 We define a country as in a recession if its real GDP growth rate is negative, and a country as in a banking crisis if there is

or was a banking crisis in the past 3 years. Banking crises indicators follow Laeven and Valencia (2020). For the sub-periods,
we break the sample up into periods around the Global Financial Crisis (GFC), in particular before 2000, from 2000-2008, and
after 2009.

vii
holdings of debt in the form of reserves by foreign central banks, particularly those in EM countries.16

For developing countries, the role of domestic investors expands considerably during recessions.

The pattern over time is also interesting. From 2000-2009, foreign investors play a larger role in picking

up debt issued by DC sovereigns, but after the GFC domestic investors’ share increases substantially as

foreign non-bank investors contract. The share of foreign official has been increasing over time.

Figure B1: Marginal Holders

1.0 1.0 1.0

0.8 0.8 0.8

0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0.0 0.0 0.0

–0.2 –0.2 –0.2

All AE EM DC No Recession No Crisis Before 2000-2009 After
recession crisis 2000 2009
Domestic Domestic
bank non-bank Domestic Foreign Foreign Foreign
central bank bank non-bank official

(a) By country group (b) Crisis and recession (c) By time period

Note: This figure plots the regression coefficients in Equation (2) for all countries under different circumstances. Panel (a)
depicts the coefficients for each investor group by country group as reported in Table 1 in the text. Panel (b) shows the
coefficients for each investor group during recession and non-recession times, and during crisis and non-crisis times. A
recession is defined by a negative real GDP growth rate. A crisis is identified following Laeven and Valencia (2020), which
includes a banking crisis, a currency crisis, and a debt crisis. Panel (c) shows the coefficients for each investor group in three
subsamples: pre-2000, 2000-2009, and post-2009.

The basic finding that Non-bank investors are important marginal investors continues to hold across

these different periods. The next subsections examine these relationships in more detail across investor

groups.

B.1.1 Recessions and Non Recessions

Table B1 reports the regression results depicted in Figures 2 for recessions and non-recessions.

16 If
these data were extended through 2020, we might see a much higher marginal share for the domestic central bank, as
some in AEs purchased amounts roughly equal to the net issuance of debt during that year (see the IMF Fiscal Monitor).

viii
 Figure B2: Marginal Holders

1.0 1.0
1.0
0.8 0.8

0.6 0.6
0.5
0.4 0.4

0.2 0.2
0.0
0.0 0.0

–0.5 –0.2 –0.2

All

recession
Recession

crisis
Crisis

2000-2009

2009

All

recession
Recession

crisis
Crisis

2000
2000-2009

2009

All

recession
Recession

crisis
Crisis

2000
2000-2009

2009
No

No

After

No

No

Before

After

No

No

Before

After
Domestic Domestic Domestic Foreign Foreign Foreign
bank non-bank central bank bank non-bank Official

(a) Advanced economies (b) Emerging markets (c) Developing countries

Note: This figure plots the regression coefficients in Equation (2) under different circumstances for advanced economies,
emerging market economies, and developing countries in Panels (a), (b) and (c), respectively. A recession is defined by a
negative real GDP growth rate. A crisis is identified following Laeven and Valencia (2020), which includes a banking crisis, a
currency crisis, and a debt crisis. The regression estimates are in Tables B1, Table B2, and Table B3.

ix
 Table B1: Marginal Holders of Sovereign Debt: Recession and No Recession

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

Dom For DomBK DomNB DomCB ForBK ForNB ForCB
Panel A: Recessions

All 0.83*** 0.17*** 0.08 0.75*** 0.00 -0.05 0.19** 0.03

(0.05) (0.05) (0.05) (0.09) (0.01) (0.04) (0.09) (0.03)

AE 0.68** 0.32 0.01 0.63** 0.05* -0.10 0.70*** -0.28

(0.26) (0.26) (0.09) (0.23) (0.02) (0.11) (0.21) (0.20)

EM 0.82*** 0.18 -0.02 0.87*** -0.03 -0.02 0.14 0.05

(0.13) (0.13) (0.06) (0.16) (0.02) (0.04) (0.14) (0.05)

DC 0.67*** 0.33* 0.07 0.46* 0.14 -0.01 0.19** 0.15**

(0.18) (0.18) (0.10) (0.22) (0.09) (0.08) (0.08) (0.06)

Panel B: Non Recessions

All 0.60*** 0.40*** 0.17*** 0.37*** 0.05*** 0.05*** 0.30*** 0.05***

(0.05) (0.05) (0.04) (0.06) (0.02) (0.02) (0.05) (0.01)

AE 0.54*** 0.46*** 0.02 0.41*** 0.11 0.05** 0.34*** 0.07**

(0.11) (0.11) (0.03) (0.09) (0.07) (0.02) (0.10) (0.03)

EM 0.67*** 0.33*** 0.25*** 0.37*** 0.05** 0.07*** 0.24*** 0.01

(0.05) (0.05) (0.05) (0.08) (0.02) (0.02) (0.06) (0.01)

DC 0.50*** 0.50*** 0.10*** 0.35*** 0.05*** 0.02 0.38*** 0.11***

(0.09) (0.09) (0.02) (0.08) (0.02) (0.02) (0.09) (0.03)

Note: This table reports the regression coefficients for Equation (2) for each investor group during recessions (Panel A) and non recessions
(Panel B). A recession is defined as a negative real GDP growth rate. Country and year FEs are included, and standard errors are clustered at
the country level and reported in the parentheses. Columns (1) and (2) represent domestic and foreign investors, respectively. Columns (3)
through (8) correspond to the six investor groups described in the text.

x
B.1.2 Banking Crisis and No Banking Crisis

Table B2 reports the regression results depicted in Figures 2 for banking crises.

Table B2: Marginal Holders of Sovereign Debt: Banking Crisis and No Banking Crisis

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

Dom For DomBK DomNB DomCB ForBK ForNB ForCB
Panel A: Banking Crisis

All 0.69*** 0.31*** 0.06 0.56*** 0.07 0.02 0.21** 0.09

(0.05) (0.05) (0.05) (0.07) (0.06) (0.02) (0.09) (0.06)

AE 0.72*** 0.28 -0.09* 0.52*** 0.29** 0.08* 0.17 0.03

(0.19) (0.19) (0.05) (0.12) (0.11) (0.04) (0.25) (0.05)

EM 0.65*** 0.35*** 0.09* 0.56*** 0.01 0.00 0.26** 0.09

(0.07) (0.07) (0.04) (0.07) (0.01) (0.02) (0.10) (0.07)

DC 0.46*** 0.54*** 0.13* 0.31** 0.02 -0.00 0.42** 0.12*

(0.12) (0.12) (0.06) (0.13) (0.02) (0.01) (0.17) (0.07)

Panel B: No Banking Crisis

All 0.59*** 0.41*** 0.18*** 0.37*** 0.04*** 0.05*** 0.32*** 0.04***

(0.05) (0.05) (0.04) (0.06) (0.01) (0.01) (0.05) (0.01)

AE 0.47*** 0.53*** 0.04* 0.41*** 0.02** 0.06* 0.46*** 0.01

(0.12) (0.12) (0.02) (0.11) (0.01) (0.03) (0.11) (0.03)

EM 0.67*** 0.33*** 0.24*** 0.39*** 0.04** 0.06*** 0.25*** 0.01

(0.06) (0.06) (0.05) (0.08) (0.02) (0.02) (0.05) (0.01)

DC 0.49*** 0.51*** 0.11*** 0.34*** 0.05** 0.01 0.40*** 0.09***

(0.11) (0.11) (0.04) (0.10) (0.02) (0.02) (0.11) (0.02)

Note: This table reports the regression coefficients for Equation (2) for each investor group during times with banking crises (Panel A) and
without banking crisis (Panel B). A country-year (n, t) observation is defined as if country n experienced a banking crisis in either of year
t, t − 1, t − 2, t − 3. Banking crisis definitions follow Laeven and Valencia (2020). For the developing countries, the definition of crisis
includes not only banking crisis, but also debt crisis and currency crisis. Country and year FEs are included, and standard errors are
clustered at the country level and reported in the parentheses. Columns (1) and (2) represent domestic and foreign investors, respectively.
Columns (3) through (8) correspond to the six investor groups described in the text.

B.1.3 Different Subperiods

Table B3 reports the regression results depicted in Figures 2 for different subperiods.

xi
B.2 Marginal Investors Considering Currency Valuation Effects

This section provides an analysis of the marginal holders of sovereign debt taking the currency valua-

tion effect into consideration.

To see the potential impact of currency, we define the impact of currency valuation on holdings

of sovereign debt n by investor group i as CVi,t (n). Then we can rewrite the general relationship for

holdings as:

Hi,t (n) − Hi,t−1 (n) = ∆ H̃i,t (n) − CVi,t (n) (B.1)

where ∆ H̃i,t (n) is the change in holdings excluding currency valuation effects.

To compute ∆ H̃i,t (n), we make two assumptions. First, all domestic investors holdings of their own

sovereign debt is denominated in local currency. Second, the local currency share of foreign investors

may be proxied by the share of local currency among bonds issued in international financial markets,

derived from the BIS international debt securities statistics. In the absence of data on the currency

breakdown by investor group, we treat this share as applying equally across all foreign investors.

Under these assumptions, we then calculate the currency valuation adjustment as:

S t ( n ) − S t −1 ( n )
CVi,t (n) = Hi,t−1 (n) × LCi,t−1 (n) ×
S t −1 ( n )

where LCi,t (n) is country n local currency share of debt investor group i’s holding of country n’s debt,

and St (n) is the price of currency n in terms of dollar, both at time t. We define the currency valuation

adjusted change in total debt as ∆ D̃t (n) = ∑iI=1 ∆ H̃i,t (n). The regression Equation (2) is in turn written

as

∆ H̃i,t (n) ∆ D̃t (n)

= ãi + ã(n) + ãt + ei,t (n), ∀i (B.2)
Dt − 1 ( n ) Dt − 1 ( n )

B.2.1 Unconditional Estimates

We first report the results using the Currency Valuation (CV) measure over all periods. These are

reported in Table B4 below. As the results show, Non-banks continue to absorb the largest proportion

of debt on the margin. Moreover, the following subsections show that these patterns continue to hold

when decomposed by recessions and non-recessions in Table B5, by banking crisis in Table B6, and by

different subperiods in Table B7.

xii
 Table B3: Marginal Holders of Sovereign Debt: Different Subperiods

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

Dom For DomBK DomNB DomCB ForBK ForNB ForCB
Panel A: Before 2000
All 0.57*** 0.43** 0.11* 0.41** 0.05* -0.01 0.42** 0.02
(0.18) (0.18) (0.06) (0.18) (0.02) (0.03) (0.16) (0.01)

EM 0.66** 0.34 0.06 0.50* 0.11*** 0.03 0.31 -0.01

(0.25) (0.25) (0.04) (0.25) (0.03) (0.04) (0.27) (0.02)

DC 0.52* 0.48* 0.14* 0.35 0.02 -0.04 0.50** 0.02*

(0.25) (0.25) (0.08) (0.25) (0.02) (0.04) (0.22) (0.01)

Panel B: 2000-2009
All 0.52*** 0.48*** 0.15** 0.32*** 0.05*** 0.03 0.40*** 0.05**
(0.07) (0.07) (0.06) (0.08) (0.02) (0.02) (0.08) (0.02)

AE 0.48*** 0.52*** -0.01 0.38*** 0.11 0.04* 0.46** 0.02

(0.17) (0.17) (0.03) (0.13) (0.08) (0.02) (0.18) (0.01)

EM 0.64*** 0.36*** 0.21** 0.40*** 0.03* 0.04 0.28*** 0.04

(0.08) (0.08) (0.08) (0.12) (0.02) (0.04) (0.08) (0.03)

DC 0.34** 0.66*** 0.12 0.18* 0.04* 0.00 0.57*** 0.09**

(0.12) (0.12) (0.07) (0.10) (0.02) (0.01) (0.15) (0.04)

Panel C: After 2009

All 0.72*** 0.28*** 0.17*** 0.50*** 0.05* 0.06*** 0.17*** 0.05**
(0.04) (0.04) (0.04) (0.08) (0.03) (0.01) (0.03) (0.02)

AE 0.61*** 0.39*** 0.11* 0.47*** 0.03 0.06* 0.22** 0.11

(0.10) (0.10) (0.05) (0.09) (0.02) (0.03) (0.08) (0.09)

EM 0.73*** 0.27*** 0.23*** 0.46*** 0.05 0.07*** 0.19*** 0.01

(0.05) (0.05) (0.05) (0.10) (0.03) (0.01) (0.05) (0.01)

DC 0.73*** 0.27** 0.04 0.60*** 0.09 0.04 0.11* 0.12**

(0.10) (0.10) (0.04) (0.16) (0.06) (0.04) (0.06) (0.05)

Note: This table reports the regression coefficients for Equation (2) for each investor group before 2000 (Panel A), 2000-2009 (Panel B), and
after 2009 (Panel C). Country and year FEs are included, and standard errors are clustered at the country level and reported in the
parentheses. Columns (1) and (2) represent domestic and foreign investors, respectively. Columns (3) through (8) correspond to the six
investor groups described in the text.

xiii
 Table B4: Marginal Holders of Sovereign Debt Conditioning on Currency

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

Dom For DomBK DomNB DomCB ForBK ForNB ForCB
All 0.70∗∗∗ 0.30∗∗∗ 0.17∗∗∗ 0.46∗∗∗ 0.07∗∗∗ 0.03∗∗∗ 0.23∗∗∗ 0.04∗∗∗
(0.05) (0.05) (0.03) (0.05) (0.02) (0.01) (0.04) (0.01)

AE 0.60∗∗∗ 0.40∗∗∗ 0.05∗∗ 0.46∗∗∗ 0.09 0.06∗∗ 0.31∗∗ 0.03

(0.12) (0.12) (0.02) (0.10) (0.06) (0.02) (0.12) (0.02)

EM 0.76∗∗∗ 0.24∗∗∗ 0.21∗∗∗ 0.47∗∗∗ 0.07∗∗∗ 0.04∗∗ 0.18∗∗∗ 0.03

(0.05) (0.05) (0.04) (0.06) (0.02) (0.01) (0.04) (0.02)

DC 0.60∗∗∗ 0.40∗∗∗ 0.11∗∗∗ 0.43∗∗∗ 0.05∗∗∗ 0.01 0.31∗∗∗ 0.07∗∗∗

(0.10) (0.10) (0.04) (0.09) (0.01) (0.01) (0.10) (0.02)

Note: Panel A of the table reports the regression coefficients for Equation (B.2) for each investor group. The first two columns represent
domestic and foreign investors, respectively. Columns (3) through (8) correspond to the six investor groups. Standard errors clustered at the
country level are reported in the parentheses. Panel B of the table reports the average share of holding by each investor group.

xiv
B.2.2 Recessions and Non-Recessions with Currency Valuation

Table B5 report the results of estimating Equation (B.2) conditioned on recessions or no recessions.

Table B5: Marginal Holders Conditioning on Currency: Recession and Non Recession

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

Dom For DomBK DomNB DomCB ForBK ForNB ForCB
Panel A: Recessions

All 0.89∗∗∗ 0.11∗∗ 0.17∗∗∗ 0.63∗∗∗ 0.09∗∗∗ -0.01 0.11∗∗ 0.01

(0.05) (0.05) (0.04) (0.05) (0.03) (0.02) (0.05) (0.02)

AE 0.86∗∗∗ 0.14 0.01 0.81∗∗∗ 0.04 -0.08 0.40 -0.18

(0.25) (0.25) (0.08) (0.24) (0.02) (0.08) (0.27) (0.15)

EM 0.85∗∗∗ 0.15∗ 0.16∗∗∗ 0.56∗∗∗ 0.13∗∗ 0.02 0.09 0.03

(0.07) (0.07) (0.04) (0.07) (0.04) (0.02) (0.07) (0.03)

DC 0.82∗∗∗ 0.18 -0.07 0.78∗∗∗ 0.10 -0.03 0.13 0.08

(0.17) (0.17) (0.09) (0.23) (0.07) (0.08) (0.10) (0.08)

Panel B: Non Recessions

All 0.67∗∗∗ 0.33∗∗∗ 0.17∗∗∗ 0.44∗∗∗ 0.06∗∗∗ 0.05∗∗∗ 0.24∗∗∗ 0.04∗∗∗

(0.05) (0.05) (0.03) (0.06) (0.01) (0.01) (0.05) (0.01)

AE 0.62∗∗∗ 0.38∗∗∗ 0.07∗∗∗ 0.45∗∗∗ 0.10 0.05∗∗ 0.28∗∗ 0.05∗

(0.11) (0.11) (0.02) (0.09) (0.07) (0.02) (0.10) (0.03)

EM 0.73∗∗∗ 0.27∗∗∗ 0.22∗∗∗ 0.45∗∗∗ 0.06∗∗∗ 0.06∗∗∗ 0.19∗∗∗ 0.01∗

(0.06) (0.06) (0.05) (0.09) (0.02) (0.02) (0.05) (0.01)

DC 0.60∗∗∗ 0.40∗∗∗ 0.12∗∗∗ 0.42∗∗∗ 0.05∗∗∗ 0.02 0.30∗∗∗ 0.08∗∗∗

(0.10) (0.10) (0.04) (0.08) (0.02) (0.01) (0.09) (0.02)

Note: This table reports the regression coefficients for Equation (B.2) for each investor group during recessions (Panel A) and non recessions
(Panel B). A recession is defined as a negative real GDP growth rate. Country and year FEs are included, and standard errors are clustered at
the country level and reported in the parentheses. Columns (1) and (2) represent domestic and foreign investors, respectively. Columns (3)
through (8) correspond to the six investor groups.

xv
B.2.3 Banking Crisis and No Banking Crisis with Currency Valuation

Table B6 reports the results of estimating Equation (B.2) conditioned on banking crisis or no banking

crisis.
Table B6: Marginal Holders Conditioning on Currency: Banking Crisis and No Banking Crisis

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

Dom For DomBK DomNB DomCB ForBK ForNB ForCB
Panel A: Banking Crisis

All 0.83∗∗∗ 0.17∗∗ 0.13∗∗∗ 0.58∗∗∗ 0.12∗∗∗ 0.00 0.12∗ 0.05

(0.07) (0.07) (0.04) (0.04) (0.04) (0.02) (0.07) (0.04)

AE 0.85∗∗∗ 0.15 0.06∗∗ 0.53∗∗∗ 0.26∗∗∗ 0.07∗∗ 0.04 0.03

(0.13) (0.13) (0.03) (0.07) (0.07) (0.03) (0.16) (0.03)

EM 0.82∗∗∗ 0.18 0.16∗∗ 0.54∗∗∗ 0.13∗∗∗ -0.01 0.15 0.04

(0.11) (0.11) (0.06) (0.06) (0.04) (0.01) (0.10) (0.04)

DC 0.65∗∗∗ 0.35∗∗ 0.05∗∗∗ 0.59∗∗∗ 0.02 0.01 0.25∗ 0.08∗∗

(0.12) (0.12) (0.01) (0.12) (0.02) (0.01) (0.13) (0.04)

Panel B: No Banking Crisis

All 0.67∗∗∗ 0.33∗∗∗ 0.18∗∗∗ 0.44∗∗∗ 0.05∗∗∗ 0.04∗∗∗ 0.25∗∗∗ 0.03∗∗∗

(0.05) (0.05) (0.03) (0.06) (0.01) (0.01) (0.05) (0.01)

AE 0.53∗∗∗ 0.47∗∗∗ 0.03 0.48∗∗∗ 0.01∗ 0.07∗∗ 0.40∗∗∗ 0.01

(0.12) (0.12) (0.02) (0.12) (0.01) (0.03) (0.11) (0.03)

EM 0.73∗∗∗ 0.27∗∗∗ 0.22∗∗∗ 0.45∗∗∗ 0.06∗∗∗ 0.05∗∗∗ 0.20∗∗∗ 0.02∗

(0.05) (0.05) (0.04) (0.08) (0.02) (0.02) (0.05) (0.01)

DC 0.57∗∗∗ 0.43∗∗∗ 0.13∗∗ 0.39∗∗∗ 0.05∗∗∗ 0.01 0.34∗∗∗ 0.08∗∗∗

(0.12) (0.12) (0.05) (0.10) (0.01) (0.02) (0.12) (0.02)

Note: This table reports the regression coefficients for Equation (B.2) for each investor group with during times with banking crises (Panel
A) and without banking crisis (Panel B). A country-year n, t observation is defined as if country n experienced a banking crisis in either of
year t, t − 1, t − 2, t − 3. Banking crisis definitions follow Laeven and Valencia (2020). For the developing countries, the definition of crisis
includes not only banking crisis, but also debt crisis and currency crisis. Country and year FEs are included, and standard errors are
clustered at the country level and reported in the parentheses. Columns (1) and (2) represent domestic and foreign investors, respectively.
Columns (3) through (8) correspond to the six investor groups in the text.

xvi
B.2.4 Different Subperiods with Currency Valuation

Table B7 reports the results of estimating Equation (B.2) over different subperiods.

Table B7: Marginal Holders Conditioning on Currency: Different Subperiods

Dom For DomBK DomNB DomCB ForBK ForNB ForCB

Panel A: Before 2000
All 0.60∗∗∗ 0.40∗∗ 0.07 ∗∗ 0.47∗∗∗ 0.06∗ -0.01 0.41∗∗ 0.01
(0.16) (0.16) (0.03) (0.17) (0.03) (0.03) (0.15) (0.01)

EM 0.63∗∗ 0.37 0.10∗ 0.42∗ 0.11∗∗∗ 0.03 0.36 -0.01

(0.22) (0.22) (0.05) (0.24) (0.04) (0.03) (0.24) (0.02)

DC 0.57∗∗ 0.43∗ 0.06∗ 0.48∗ 0.02 -0.04 0.44∗ 0.03∗

(0.24) (0.24) (0.03) (0.23) (0.02) (0.03) (0.21) (0.01)
Panel B: 2000-2009
All 0.65∗∗∗ 0.35∗∗∗ 0.18∗∗∗ 0.41∗∗∗ 0.06∗∗∗ 0.03∗ 0.28∗∗∗ 0.04∗∗
(0.07) (0.07) (0.04) (0.07) (0.02) (0.02) (0.07) (0.02)

AE 0.58∗∗∗ 0.42∗∗ 0.03 0.46∗∗∗ 0.10 0.05∗∗ 0.36∗∗ 0.01

(0.16) (0.16) (0.02) (0.11) (0.07) (0.02) (0.17) (0.01)
EM 0.73∗∗∗ 0.27∗∗∗ 0.23∗∗∗ 0.44∗∗∗ 0.06∗∗ 0.03 0.20∗∗∗ 0.03
(0.08) (0.08) (0.06) (0.09) (0.03) (0.03) (0.07) (0.02)
DC 0.52∗∗∗ 0.48∗∗∗ 0.15∗∗ 0.33∗∗∗ 0.04∗∗ 0.01 0.41∗∗ 0.07∗
(0.16) (0.16) (0.06) (0.11) (0.01) (0.01) (0.17) (0.03)

Panel C: After 2009

All 0.80∗∗∗ 0.20∗∗∗ 0.16∗∗∗ 0.56∗∗∗ 0.07∗∗∗ 0.04∗∗∗ 0.13∗∗∗ 0.03∗∗
(0.05) (0.05) (0.04) (0.07) (0.02) (0.01) (0.03) (0.01)

AE 0.68∗∗∗ 0.32∗∗∗ 0.13∗ 0.55∗∗∗ 0.01 0.05 0.17∗ 0.10

(0.11) (0.11) (0.07) (0.11) (0.02) (0.03) (0.09) (0.08)

EM 0.82∗∗∗ 0.18∗∗∗ 0.20∗∗∗ 0.54∗∗∗ 0.08∗∗∗ 0.04∗∗∗ 0.13∗∗∗ 0.01

(0.06) (0.06) (0.04) (0.09) (0.03) (0.02) (0.05) (0.01)

DC 0.76∗∗∗ 0.24∗∗ 0.04 0.64∗∗∗ 0.08 0.05 0.10∗ 0.09∗

(0.09) (0.09) (0.03) (0.14) (0.05) (0.03) (0.06) (0.05)

Note: This table reports the regression coefficients for Equation (B.2) for each investor group before 2000 (Panel A), 2000-2009 (Panel B), and
after 2009 (Panel C). Country and year FEs are included, and standard errors are clustered at the country level and reported in the
parentheses. Columns (1) and (2) represent domestic and foreign investors, respectively. Columns (3) through (8) correspond to the six
investor groups.

xvii
B.3 Non-bank Regressions: Euro Area SHS, US Treasuries, and UK Gilts

In this subsection, we report the results of estimating the marginal investor decomposition Equation (2)

for a disaggregated data set of Euro Area investors, US Treasuries, and UK Gilts.

B.3.1 Euro Area Data

Table B8 reports the estimates of the share of marginal Euro Area investor holdings of non-Euro Area

sovereign investments, as depicted in Figure 3 in the text.

Table B8: Marginal Foreign Non-bank Investors - Euro Area

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

Gov Banks non-bank HH InsurPens NFC OthFin
All 0.00*** 0.23*** 0.77*** 0.01 0.14** 0.00* 0.85***
(0.00) (0.07) (0.07) (0.01) (0.06) (0.00) (0.12)

AE 0.00 0.53*** 0.47*** 0.02 0.12* 0.01 0.85**

(0.00) (0.14) (0.14) (0.01) (0.06) (0.01) (0.29)

EM 0.00*** 0.10* 0.90*** 0.01 0.14* 0.00 0.85***

(0.00) (0.05) (0.05) (0.01) (0.07) (0.00) (0.12)

Note: This table reports regression coefficients of Equation (2) on sovereign debt issued by non-European countries for
different investor groups within Europe. Columns (1) through (3) report coefficients for governments, banks and the
non-bank sector. Columns (4) through (7) disaggregates non-banks into households (HH), insurance and pension funds
(insurPens), non-financial corporations (NFC), and other financial institutions (OthFin). Standard errors clustered at country
and year level are reported in the parentheses.

In particular, in the row labeled ”All”, this table reports the estimates of Equation (2) treating the

denominator as the total holdings of Non-Euro Area sovereign debt by Euro Area investors. Similarly,

in the rows labeled ”AE” and ”EM”, this denominator and related holdings breakdowns are based on

total holdings by EA investors of AE and EM debt, respectively.

B.3.2 US Treasuries

Table B9 gives the estimates of Equation (2) for the US Treasuries using TIC data. In this case, for the

”All” row, the denominator is given by total holdings of US Treasuries while the numerator corresponds

to holdings by the reported investor groups. In the row labeled ”Non-bank”, the denominator is the

total holdings of Nonbank investors while the numerator is the subgroup of Nonbank investors given

by Money Market Funds (MMFs), Hedge Funds and Households (HF/HH), Insurance and Pensions

(IP), Other Financials (OthFin) and Non Financials (NonFin).As the table shows, non-bank investors

xviii
are particularly important within the domestic investor group. Taken together, Money Market funds

and Hedge funds/households comprise that largest component of domestic nonbanks.

xix
 Table B9: Marginal Investors in US Treasuries

Marginal Share Marginal Non-Bank Share

(1) (2) (3) (4) (5) (6) (7) (8) (9)
Foreign DomCB DomBank DomNB MMFs HF/HH I&P OthFin NonFin
All 0.247∗∗∗ 0.132∗∗∗ 0.061∗∗∗ 0.560∗∗∗
(0.039) (0.040) (0.012) (0.049)

Non-Bank 0.241∗∗∗ 0.314∗∗∗ 0.150∗∗∗ 0.122∗∗∗ 0.173∗∗∗

(0.037) (0.051) (0.034) (0.031) (0.058)

Constant 0.004∗∗∗ 0.001 -0.001∗∗ -0.004∗∗∗ 0.000 -0.002∗∗ 0.001∗∗ 0.001∗ -0.001
(0.001) (0.001) (0.000) (0.002) (0.001) (0.001) (0.001) (0.001) (0.001)
Observations 104 104 104 104 104 104 104 104 104
Note: This table reports estimates from Equation (2) for the holders of US Treasuries over 1995q1-2020q4. “HF/HH” includes Hedge
funds, private equity, private trusts, and direct household holdings. “NonFin” is all non-financial holders excluding households. “Oth-
Fin” is all other financial institutions apart from funds and Insurance and Pension (I&P). * p < 0.10, ** p < 0.05, *** p < 0.01
xx
B.3.3 UK Gilts

Table B10 reports estimates for Equation (2) where the holdings are the shares of UK Gilts.

Table B10: Marginal Investors in UK Gilts

Marginal Share Marginal Non-Bank Share

(1) (2) (3) (4) (5) (6) (7)
Funds
Foreign DomCB DomBank DomNB & Oth I&P NonFin

All 0.174∗∗∗ 0.256∗∗∗ 0.126∗∗∗ 0.443∗∗∗

(0.029) (0.047) (0.025) (0.047)

Non-Bank 0.510∗∗∗ 0.481∗∗∗ 0.0085∗

(0.055) (0.054) (0.004)

Constant 0.003∗∗ -0.000 -0.002∗∗ -0.001 -0.003∗∗ 0.003∗∗ -0.000

(0.001) (0.002) (0.001) (0.002) (0.001) (0.001) (0.000)

Observations 104 104 104 104 104 104 104

Note: This table reports estimates from Equation (2) for the holders of UK Gilts over 1995q1-2020q4. “Funds
& Oth” includes all financial firms besides banks, and insurance companies and pension funds (I&P). * p <
0.10, ** p < 0.05, *** p < 0.01

Table B10 gives the estimates of Equation (2) for the UK Gilts. In the row labeled ”All”, the de-

nominator is given by total holdings of UK Gilts while the numerator corresponds to holdings by the

reported investor groups. In the row labeled ”Non-bank”, the denominator is the total holdings of Non-

bank investors while the numerator is the subgroup of Nonbank investors given by ”Funds and Other”,

Insurance and Pensions (IP), and Non Financials (NonFin). As the table shows, non-bank investors are

particularly important within the domestic investor group.

xxi
C Supplemental Results for Instrument Construction

This appendix provides supplementary information for the instruments used in Section 3 of the text.

C.1 Reduced-form Estimates of IV Construction: EMs and Yields

This section reports the regression coefficients for the first-stage regression that constructs the instru-

ments. The following Table C11 gives the results for Equation (8) for Emerging Market countries.

Table C11: Reduced-form Holding Regression: EM

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

DomBK DomNB DomCB ForBK ForNB ForCB
GDP -0.18∗∗ 0.28∗∗∗ -0.71∗∗∗ -0.28∗ -1.74∗∗∗ -1.54∗∗∗
(0.09) (0.17) (0.32) (0.18) (0.29) (0.18)

GDP Growth -0.35 -2.22∗∗ 2.49 -3.85∗∗∗ 1.88 -1.32

(0.62) (1.12) (2.36) (1.18) (1.73) (1.20)

Inflation -2.12∗∗∗ -0.15 -1.25 -2.05∗∗∗ 1.12∗ -0.52

(0.24) (0.41) (0.78) (0.46) (0.64) (0.47)

Exp-to-GDP -0.00 -0.01 -0.43 -0.20 -0.22 0.00

(0.10) (0.18) (0.34) (0.19) (0.35) (0.20)
Observations 391 356 324 391 340 380
R2 0.98 0.95 0.88 0.92 0.88 0.90

Note: This table reports the reduced-form estimates of the first-stage regression of instrument construction for EMs, with country and year
fixed effects. The sample spans 1996-2018 at annual frequency. The dependent variable is the logarithm of the holdings to GDP by investor
group indicated in the column title to GDP. The standard errors are reported in the parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

C.2 Projected Debt-to-GDP Estimates: EMs

Table C12 reports the projections of debt-to-GDP, dt (n) for EM countries using Equation (9). As de-

scribed in the text, this variable is projected on exogenous characteristics and its lag modified by the

gross GDP growth. A coefficient of 0.72 indicates about a 28% of mean reversion in the level of debt-

to-GDP. As described in the text, this finding may be interpreted as an average maturity of 3.6 years,

which is close to the 5-year bond yield we choose in the main analysis.

xxii
 Table C12: Projected Debt-to-GDP Supply: EMs

Debt-to-GDP
Lag Debt-to-GDP 0.72∗∗∗
(0.03)

GDP Growth -0.09

(0.11)

Inflation 0.11∗∗∗
(0.04)

Exp-to-GDP -0.03∗
(0.02)

Observations 362
R2 0.96

Note: This table reports the estimates of Equation (9) for emerging market economies. The sample spans 1996-2018 at annual frequency. The
standard errors are reported in the parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

C.3 IV Construction: AE

This section reports the estimates used to construct the Advanced Economy instruments for Table 4 in

the text. The following table gives the results of Equation (8) for Advanced Economies.

C.4 Projected Debt-to-GDP Estimates: AEs

Table C14 reports the projections of debt-to-GDP, dt (n) for AE countries using Equation (9). As de-

scribed in the text, this variable is projected on exogenous characteristics and its lag modified by the

gross GDP growth.

xxiii
 Table C13: Reduced-form holding regressions: AEs

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

DomBK DomNB DomCB ForBK ForNB ForCB
GDP -0.30∗∗∗ -0.92∗∗∗ -2.10∗∗ 0.74∗∗∗ 0.66 1.09∗∗∗
(0.22) (0.31) (0.83) (0.21) (0.41) (0.22)

GDP growth 1.76 -5.90∗∗∗ 3.92 -2.15∗ -2.63 2.74∗∗

(1.23) (1.73) (4.43) (1.16) (2.19) (1.23)

Inflation -4.29∗ -1.91 -0.13 10.16∗∗∗ 0.08 5.50∗∗

(2.43) (3.44) (9.11) (2.29) (4.32) (2.43)

Exp-to-GDP 0.44 -0.37 2.72∗∗ 0.21 0.97∗ -0.20

(0.29) (0.43) (1.14) (0.28) (0.54) (0.29)
Observations 341 322 312 342 326 341
R2 0.99 0.93 0.82 0.95 0.86 0.96

Note: This table reports the reduced-form estimates of the first-stage regression of instrument construction for AEs, with country and year
fixed effects. The sample spans 1996-2018 at annual frequency. The dependent variable is the logarithm of holdings to GDP by the group
indicated in the column title to GDP. The standard errors are reported in the parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

Table C14: Projected Debt-to-GDP Supply: AEs

Lag Debt-to-GDP 0.72∗∗∗

(0.03)

GDP Growth -0.71∗∗∗

(0.23)

Inflation -0.98∗∗
(0.45)

Exp-to-GDP 0.15∗∗∗
(0.05)

Observations 323
R2 0.98

Note: This table reports the estimates of Equation (9) for advanced economies. The sample spans 1996-2018 at the annual frequency. The
standard errors are reported in the parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

xxiv
D Supplemental Results for Demand System Estimation

In this appendix, we provide more detailed results about the demand system approach of estimation in

Section 3.3 of the paper.

D.1 Predictive Regression Results

Table D15 reports the return predictive regression results for Equation (15) and (13) in the main text.

Table D15: Bond Return Predictive Regression

(1) (2)
Domestic Excess Return Foreign Excess Return

pt (n) -0.59∗∗∗ -0.75∗∗∗

(0.05) (0.08)

qt (n) 0.23∗∗∗
(0.06)

Observations 367 367

R2 0.34 0.27

As described there, pt (n) is the logarithm of the price of a 5 year zero coupon bond in local currency

for country n and qt (n) is the logarithm of its real exchange rate relative to the U.S. dollar. In particular,

qt (n) ≡ st (n) − zt (n) where st (n) is the logarithm of the US dollar price of currency n and zt (n) is the

logarithm of the ratio of the price index of country n to the US price index.

D.2 Yield Elasticity

We next describe the derivation of the investor demand elasticity following the literature of Koijen and

Yogo (2019) and Jiang et al. (2022).

D.2.1 Foreign Investors

As described above, foreign non-EM investors hold the sovereign debt of all EM countries plus an

outside asset that does not include this debt. To calculate the demand elasticity relative to expected

returns µ, differentiate with respect to expected returns the ratio of shares for a given country n given

xxv
in Equation (11) in the text. This implies:

dδi,t (n) d(ln ωi,t (n) − ln(1 − ∑kN=1 ωi,t (k ))) d ln wi,t (n) w (k ) d ln wi,t (k )
= = + ∑ i,t = λ0,i (D.1)
dµi,t (n) dµi,t (n) dµi,t (n) k
w i,t (0) dµi,t ( n )

Without loss of generality, consider n = 1 and denote κn,j ≡ (d ln wi,t (n)/dµi,t ( j)). Suppressing the

investor indicator i for parsimony, the relationship can be rewritten,

N
ω (1) ω (k)
κ11 + κ11 + ∑ κ = λ0 (D.2)
ω (0) k =2
ω (0) k,1

Multiplying both sides by ω (0), implies:

N
(ω (0) + ω (1))κ1,1 + ∑ ω (k)κk,1 = λ0 ω (0) (D.3)
k =2

Furthermore, note that for j 6= 1,

N
ω (1) ω (k)
κ j,1 + κ1,1 + ∑ κ =0 (D.4)
ω (0) k =2
ω (0) k,1

Multiplying both sides by ω (0), we get

N
ω (0)κ j,1 + ω (1)κ1,1 + ∑ ω (k)κk,1 = 0 (D.5)
k =2

From the equation above, we see that κ j,1 for j 6= 1 does not depend on j. Therefore,

ω (1)
κ j,1 = − κ1,1 for j 6= 1 (D.6)
1 − ω (1)

Substituting this variable into the above Equation (D.3) and solving for κ11 implies:

κ1,1 = λ0 (1 − ω (1)) (D.7)

Since this relationship holds for any arbitrary country n and individual investor i, then clearly the

demand elasticity for foreign investors can be written more generally as:

d ln Hi,t (n) d ln wi,t (n)

= = λ0,i (1 − ωi,t (n)) (D.8)
dµi,t (n) dµi,t (n)

xxvi
D.2.2 Domestic Investors

We derive the elasticity of demand with respect to expected excess returns for domestic investors by

differentiating Equation (12) in the text. In particular, we differentiate this share as in:

d ln( Hi,t (n)) d ln ωi,t (n)

= (D.9)
dµi,t (n) dµi,t (n)

subject to the constraint that the shares of wealth sum to one. In the case of domestic investors, we

assume they hold no other sovereign debt so that the share for domestic investors plus the share of

outside assets sum to one; that is, ωi,t (n) = 1 − ωi,t (0, n). Thus, differentiating the ratio of portfolio

shares in Equation (12) with respect to expected returns implies:

d(ln ωi,t (n) − ln(1 − ωi,t (n)) 1 d ln ωi,t (n)

== = λ0,i
dµi,t (n) 1 − ωi,t dµi,t (n)

Therefore,
d ln ωi,t (n)
= λ0,i (1 − ωi,t (n)) (D.10)
dµi,t (n)

Comparing this equation with Equation (D.8) makes clear that the form of the demand elasticities are

the same between foreign and domestic investors. Moreover, these demand elasticity measures are the

same as Koijen and Yogo (2019) and Jiang et al. (2022).

Price elasticity. Bond price and expected excess return are linked through the predictive regression of

excess returns described above. Using this relationships implies:

m (n)
d ln Hi,t
= λ0,i (1 − ωi,t (n))φ0i (D.11)
d ln Pt (n)

We lack the data to assess the response of exchange rate in response to bond price change, thus we

must indirectly assume that the change of bond price does not affect demand of debt through changing

exchange rate.

Yield elasticity.
m (n)
d ln Hi,t 1
= − λ0,i (1 − ωi,t (n))φ0i (D.12)
dyt (n) T

xxvii
E Counterfactual Derivation

This section provides the detailed algebra for the calculation of counterfactual measures.

E.1 Borrowing Cost Sensitivity

We are interested in the following question: if the government wants to increase debt by one percent,

how much will yield increase? To derive this measure, we start from the following equation:

6
Pt (n) Dt (n) = ∑ Pi,t (n) Hi,t (n) (E.1)
i =1

where Pt (n) is the price of debt issued by country n and Pi,t (n) is the price faced by investor group

i. In equilibrium, Pi,t (n) = Pt (n). Thus, the question above can be restated as: what will happen

to the yield if the government issues one more percent of (market value) debt? This question can be

answered by calculating the change in yield for a percentage change in the market value of debt; that
dyt (n)
is, ξ ≡ d ln( Pt (n) Dt (n))
.

To calculate this measure, note that Pt (n) = exp(− Tyt (n)). Therefore, we can rewrite this measure

as:
dyt (n) dyt (n)
ξ≡ = (E.2)
d ln ( Pt (n) Dt (n)) − Tdyt (n) + d ln Dt (n)

From the market clearing condition, we can express the yield as:

!
6
1
yt (n) = ln Dt (n) − ln ∑ Pti (n) Hi (n) (E.3)
T i =1

Differentiating with respect to the logarithm of debt then implies:

 
I d( Pti (n) Hi,t (n)) dHi,t (n)
dyt (n) 1 ∑
1  i =1 dHi,t (n) dDt (n) t
D ( n )
= −  (E.4)

I i
d ln Dt (n) T T ∑i=1 Pt (n) Hi,t (n)


dHi,t (n)
Recall that from Section 2, the investor group i’s marginal financing share dDt (n)
= ai . Moreover, define
∂yt (n) 1 dyt (n)
the inverse demand elasticity as m (n)
∂ ln Hi,t
= η̃i . Then with these definitions, we can re-express d ln Dt (n)

in Equation (E.4) as: !

I I
dyt (n) 1 1 ai ai
ξ̃ ≡ = − 1−T∑ = ∑ η̃i (E.5)
d ln Dt (n) T T η̃
i =1 i i =1

Empirically, our estimates in Table 3 in Panel B provide the semi-elasticity of market value of holding

xxviii
with respect to yield. That is, we can denote our empirical estimates as:

1 dyt (n) dyt (n) 1/η̃i

≡ m ( n ) = − Tdy ( n ) + d ln H ( n ) = − T/ η̃ + 1
ηi d ln Hi,t t i,t i

Therefore, substituting these estimates into Equation (E.2) implies:

ai
dyt (n) dyt (n)/d ln Dt (n) ∑iI=1 ηi + T
ξ= = = I a i ηi (E.6)
d ln ( Pt (n) Dt (n)) − Tdyt (n)/d ln Dt (n) + 1 ∑ i =1 T + ηi

as given in equation (17) in the text. ξ is the inverse of the aggregate demand semi-elasticity with

respect to yield in the EM sovereign debt market.

E.2 Sensitivity to Characteristics Changes

In this section, we derive the impact on yield implied by a change in one of the characteristics that we

denote Xt . Again we start with the market clearing condition of Equation (E.1). The market clearing

condition implicitly defines a function of Pt (n) = P( Xt (n), Dt (n)).

We first calculate how much the price of debt would change in response to a unit change in Xt (n)

given the book value of debt. Taking the first-derivative with respect to Xt implies:

I ∂H m ( n ) m
d ln Pt (n) d ln Pt (n) ∂Hi,t (n)
Dt (n) Pt (n) =∑ i,t
+ (E.7)
dXt (n) i =1
∂ ln Pt (n) dXt (n) ∂Xt (n)

Rearranging terms yields:

∂ ln H m (n)
d ln Pt (n) ∑iI=1 ∂Xt (i,tn) ψi
= ∂ ln H m (n)
(E.8)
dXt (n) 1 − ∑iI=1 ∂ ln Pi,tt (n) ψi

where ψi is the average share of debt held by investor i. We can then write the yield response to a

change in a given characteristic Xt (n) as:

I ∂ ln H m (n)
1 ∑i=1 ∂Xt (n) ψi
i,t
dyt (n)
ζ̃ = =− (E.9)
dXt (n) T 1 − I ∂ ln Hi,tm (n) ψ
∑i=1 ∂ ln Pt (n) i

Clearly, ζ̃ is the yield response to a unit change in Xt (n) given the book value of debt. But the market

value will change with the price response. Therefore, as above, we calculate how much the book value

of debt would have to change for a given market value.

xxix
 Recall that the price of debt can be written as a function of Dt (n) and Xt (n) as P( Dt (n), Xt (n)). We

can rewrite the change of market value in response to a unit change in Xt (n) as

d(ln P( Dt (n), Xt (n)) + d ln Dt (n) ∂ ln Pt (n) d ln Dt (n) ∂ ln Pt (n) d ln Dt (n)

= + + =0 (E.10)
dXt (n) ∂ ln Dt (n) dXt (n) ∂Xt (n) dXt (n)

We then solve for the percentage change in the book value of debt due to a change in the given charac-

teristic as:
d ln Dt (n) ∂ ln Pt (n)/∂Xt (n) T ζ̃
=− = (E.11)
dXt (n) 1 + ∂ ln Pt (n)/∂ ln Dt (n) 1 − T ξ̃

where ξ is the inverse of aggregate demand elasticity derived in section E.1. Then defining the yield

response to a unit change in characteristics Xt (n) while keeping the market value of debt unchanged as

ζ, this variable can be expressed as:

dyt (n) T ζ̃ ζ̃
ζ = ζ̃ + = (E.12)
d ln Dt (n) 1 − T ξ̃ 1 − T ξ̃

This expression verifies Equation (18) in the text.

E.3 Spillover of Individual Holding

In this section, we derive the impact on the yield if investor i’s latent demand increases by one percent.

We also consider how much each investor’s holdings will change. We start from the market clearing

condition of Equation (E.1). The market clearing condition implicitly defines a function of Pt (n) =

P( Xt (n), Dt (n), ε t (n)) where ε t (n) is the vector of each investor’s latent demand for debt issued by

country n.

Again, we first fix the book value of debt Dt (n) and take the first-derivative with respect to investor

k’s latent demand ε k,t (n), obtaining:

I ∂H m ( n ) m (n)
d ln Pt (n) d ln Pt ∂Hk,t
Dt (n) Pt (n) =∑ i,t
+ (E.13)
dε k,t (n) i =1
∂ ln Pt dε k,t (n) ∂ε k,t (n)

Thus, the response of price to the change in latent demand is:

m ( n ) /∂ε ( n )
d ln Pt (n) ∂Hk,t k,t ψk
= m (n) = ∂ ln H m (n)
(E.14)
dε k,t (n) I ∂Hi,t
Dt (n) Pt (n) − ∑i=1 ∂ ln Pt (n) I
1 − ∑i=1 ψi ∂ ln Pi,tt (n)

xxx
We can then write the yield response as

dyt (n) 1 ψk
ϕ̃ = =− (E.15)
dε k,t (n) T1+ 1
T ∑iI=1 ψi ηi

Next, we calculate how much the book value of debt would have to change in order to keep the market

value of debt unchanged. This calculation implies the differential relationship:

d ln( P( Dt (n), Xt (n), ε t (n)) + d ln Dt (n)) d ln Pt (n) d ln Dt (n) d ln Pt (n) d ln Dt (n)

= + + = 0 (E.16)
dε k,t (n) d ln Dt (n) dε k,t dεkt dεkt

Solving for the implied percentage change of book value of debt to keep the market value of debt

unchanged implies:
d ln Dt (n) T ϕ̃
= (E.17)
dε k,t (n) 1 − Tξ

Then, defining the object of interest as ϕ, this relationship can be expressed as:

ϕ̃
ϕ= (E.18)
1 − T ξ̃

This derivation verifies Equation (19) in the text.

xxxi
Previous volumes in this series
1098 Long term debt propagation and real Mathias Drehmann, Mikael Juselius
May 2023 reversals and Anton Korinek

1097 Dampening global financial shocks: can Katharina Bergant, Francesco

May 2023 macroprudential regulation help (more than Grigoli, Niels-Jakob Hansen and
capital controls)? Damiano Sandri

1096 Money Market Funds and the Pricing of Sebastian Doerr, Egemen Eren and
May 2023 Near-Money Assets Semyon Malamud

1095 Sectoral shocks, reallocation, and labor Joaquín García-Cabo, Anna Lipińska
April 2023 market policies and Gastón Navarro

1094 The foreign exchange market Alain Chaboud, Dagfinn Rime and
April 2023 Vladyslav Sushko

1093 Sovereign risk and bank lending: evidence Yusuf Soner Baṣkaya, Bryan Hardy,
April 2023 from the 1999 Turkish earthquake Ṣebnem Kalemli-Özcan and Vivian
Yue

1092 Mobile Payments and Interoperability: Milo Bianchi, Matthieu Bouvard,

April 2023 insights from the academic literature Renato Gomes, Andrew Rhodes
and Vatsala Shreeti

1091 FinTech, investor sophistication and financial Leonardo Gambacorta, Romina

April 2023 portfolio choices Gambacorta and Roxana Mihet

1090 Tackling the fiscal policy-financial stability Claudio Borio, Marc Farag and
April 2023 nexus Fabrizio Zampolli

1089 Intraday liquidity around the world Biliana Alexandrova Kabadjova,

April 2023 Anton Badev, Saulo Benchimol
Bastos, Evangelos Benos, Freddy
Cepeda-Lopéz, James Chapman,
Martin Diehl, Ioana Duca-Radu,
Rodney Garratt, Ronald Heijmans,
Anneke Kosse, Antoine Martin,
Thomas Nellen, Thomas Nilsson, Jan
Paulick, Andrei Pustelnikov,
Francisco Rivadeneyra, Mario Rubem
do Coutto Bastos and Sara Testi

1088 Big techs and the credit channel of Fiorella De Fiore, Leonardo
April 2023 monetary policy Gambacorta and Cristina Manea

1087 Crypto carry Maik Schmeling, Andreas

April 2023 Schrimpf and Karamfil Todorov

All volumes are available on our website www.bis.org.