Mapping Cross-Border Financial Linkages - A Supporting Case for Global Financial Safety Nets

International Monetary Fund

doi:10.5089/9781498338806.007.A001

Jump to Content Jump to Main Navigation

Mapping Cross-Border Financial Linkages - A Supporting Case for Global Financial Safety Nets

Author: International Monetary Fund

Publication Date:: 06 Jan 2011

eISBN:: 9781498338806

Language:: English

Keywords:: PP; IMF staff calculation; asset class; global financial system; bank claim; absorptive capacity; bank assets correspond; cross-border holding; asset price response; assets corresponds to a shift; market capitalization

Download PDF (12 MB)

This paper maps cross-border financial linkages and identifies factors that drive them, contributing to the discussion on the appropriate design of a global financial safety net (GFSN). It builds on previous staff work and complements the findings of the companion paper on the Analytics of Systemic Crises and the Role of Global Financial Safety Nets. This paper notes the growing roles of financial linkages and complexity in injecting latent instability into the global financial system, underscoring the value of a GFSN design that is effective in forestalling the risk that a localized liquidity shock propagates through the global financial network turning into a large-scale systemic crisis.

Abstract

I. Executive Summary

Objectives. This paper maps cross-border financial linkages and identifies factors that drive them, contributing to the discussion on the appropriate design of a global financial safety net (GFSN). It builds on previous staff work and complements the findings of the companion paper on the Analytics of Systemic Crises and the Role of Global Financial Safety Nets. This paper notes the growing roles of financial linkages and complexity in injecting latent instability into the global financial system, underscoring the value of a GFSN design that is effective in forestalling the risk that a localized liquidity shock propagates through the global financial network turning into a large-scale systemic crisis.

Mapping the linkages. Cross-border financial linkages have increased dramatically over time and have become more complex. Yet, a few “core” advanced economies (AEs), including some financial centers, still dominate the web of linkages across asset classes and regions, both as sources and recipients. As a result, emerging markets’ (EMs) strongest linkages remain with AEs, even though cross-EM linkages have increased very rapidly during the last decade (from a low base).

Systemic instability. Increased cross-border financial linkages promote risk diversification at the individual country level, reducing exposure to localized shocks. However, increased interconnectedness, by facilitating transmission of shocks, also generates a network externality that makes the global financial network more prone to systemic risk—the risk that shocks to a “core” node leads to a breakdown of the entire network. Moreover, as the extent and complexity of cross-border financial linkages grow, investor information about specific exposures becomes less certain, amplifying systemic risks from panic responses to shocks.

Shock transmission. The paper points out that (i) countries with shallow domestic financial markets and concentrated exposures to a few lenders are more prone to synchronized shifts in cross-border flows; and (ii) common factors (such as global risk aversion) increasingly drive global financial markets and tend to intensify abruptly during periods of stress, amplifying shock transmission. These features point to potentially large costs of systemic shocks to “crisis bystanders” (countries with relatively strong fundamentals for which the likelihood of an idiosyncratic crisis is normally low), and reinforce the case for a GFSN that is designed to help ring-fence such countries from systemic shock contagion.

Determinants of linkages. Empirical evidence shows that geographical and historical factors remain important determinant of cross-border linkages—in particular, stronger linkages occur among economies closer to each other, and those that are larger, more developed, and financially more advanced. Beyond providing general principles that could underpin the design of a GSFN, these findings suggest that an insurance mechanism against sudden shifts in cross-border exposures driven by aggregate or global shocks is essential to complement local or regional risk-sharing mechanisms.

II. Context and Motivation¹

1. Global trends. Global economic linkages have intensified dramatically over the past two decades, underpinned by an exponential rise in trade and financial flows (Figure 1). Cross-border linkages have been dominated by financial flows among advanced economies (AEs). However, flows to, and among, emerging markets (EMs) have also risen in importance, both in absolute terms and in relation to the size of their economies.² As linkages among economies have intensified, their patterns have also grown in complexity. One example of growing complexity in cross-border financial networks is the thickening of the web of financial links among European EMs observed during the last decade (Figure 2).

2. Focus of this paper. Understanding the evolving nature of cross-border financial linkages—the “plumbing” of the global economy—helps map out the channels through which shocks are potentially transmitted, and better tailor global and local policy responses to shocks. The recent crisis has vividly illustrated the costs and benefits of increased interconnectedness—“linkages” and “interconnectedness” will be used interchangeably in this paper—exposing lacunae in the global financial architecture. This experience carries important lessons, which this paper and the companion paper on the Analytics of Systemic Crises and the Role of Global Financial Safety Nets (Systemic Crises paper from now) aim to internalize. A key goal of this paper is to map and explain the drivers of cross-border financial linkages and their macroeconomic consequences. The resulting empirical evidence lends support to the establishment and appropriate design of a global financial safety net (GFSN) that mitigates the impact of global liquidity shocks.

3. Linkages and systemic instability. This paper’s main thesis is that efforts to increase the resilience of the global financial system face a delicate trade-off between (a) the benefits at the country level from increased international risk diversification—a force that pushes toward increasing the system’s overall interconnectedness (i.e., the number of linkages in the network)—and (b) the increased systemic risk that this heightened interconnectedness generates at the global level. This trade-off can be understood using insights from network theory, which has been applied to understand the properties of complex financial networks and their implications for financial stability (Box 1 and Annex II). It is the fundamental—and to some extent unavoidable—tension between country-level benefits and the externality they create via increased systemic fragility that provides the key rationale for erecting truly global defenses against liquidity shocks that go beyond the national and regional boundaries.

Box 1.

Financial Networks and Systemic Risks³

Robust yet fragile network. As argued in the financial network literature, interconnectedness is a double-edged sword: it has the potential of making a network more robust via improved risk sharing, yet it could also render a network more fragile by increasing systemic risk. At a low level of interconnectedness, additional links among countries make the system more “robust” by strengthening individual countries’ ability to withstand shocks. Efforts to reduce interconnectedness may reduce welfare from the perspective of an individual country. However, in a complex network, even an initially localized shock to a “core node” could propagate widely and in a non-linear fashion through the network, leading to costly systemic crises. Such latent fragility of complex networks results from the tension between country-level optimal choices (better risk sharing via more linkages) and the externalities to the system (higher systemic risks). These negative externalities are unlikely to be internalized by individual countries.

Complex networks and incomplete information. A complex network is likely to give rise to incomplete information and increase the potential for herding behavior, flight to quality and liquidity crunches, as shown in a growing body of literature. As a result, even countries not considered to be “systemic” ex ante (i.e., small and less connected countries) could trigger “systemic” market responses, as their crises serve as a “wake up” call to creditors, triggering a broad-based pull back from other similarly-situated countries.

Concentration risks. Additional risks arise in a network where recipient countries have an unusually large concentration of exposures to a few sources. In these circumstances, a large shock hitting a main source could create more severe deleveraging than in an alternative situation where all links are uniform across sources. Higher concentration increases exposure to local/regional shocks and thus reduces the benefit of regional or local insurance mechanism based on local/regional risk-sharing.

Stylized facts. Based on measures developed in the literature to capture network characteristics (Annex II), interconnectedness is much stronger for AEs than EMs and has generally increased in the last decade across the two groupings as well as across asset classes. In EMs, interconnectedness is higher for cross-border bank claims than portfolio claims. Concentration is still significantly higher in EMs than AEs, but has generally been on a declining trend in the past decade. The most notable exception is cross-border bank claims in European EMs, for which concentration increased rapidly in the years before the crisis.

4. National defenses. The increased systemic instability arising from growing cross-border linkages often manifests itself in capital flow volatility. This volatility can be mitigated in principle at the national level by accumulating international reserves as a form of self insurance and taxing the externality-generating flows (or by throwing sand in the wheels through administrative or prudential measures). Taxing away the negative externality is, however, complicated by the difficulty of measuring the unobservable externality and by the need to properly account for the (equally difficult to quantify) multilateral benefits of increased interconnectedness. Pursuing self insurance is also constrained by the potential fiscal costs of carrying (low-yielding) foreign assets, the diminishing returns to reserve accumulation (IMF, 2010d), and the fear of using reserves in times of crisis. Not surprisingly, the recourse to taxes and self insurance has varied considerably across countries (IMF, 2011c, and Magud and others, 2011).

5. Regional and global defenses. To the extent that national defenses are insufficient to reduce the source of volatility and instability embedded in the global financial system, regional and global financing mechanisms have a role to play in cushioning the impact of this residual volatility on individual countries and on the system as a whole (see IMF, 2011d). From this perspective, the recent global crisis underscores the value of an effective global mechanism to coordinate liquidity injections and other policy responses: in 2008–09, a number of “crisis bystanders”—countries with relatively strong fundamentals (IMF, 2010d, and the Systemic Crises paper)—were hit later and less severely than countries with weaker fundamentals and were able to recover from the crisis more rapidly; however, they still suffered a deep output contraction (Figure 3).⁴ This, combined with the limited opportunities to diversify against the risk of cross-border shocks (see below), points to the significant global welfare gains from an effective global financial safety net.

6. Country coverage. To illustrate the benefits of a GFSN, the empirical analysis of this paper focuses mostly, but not exclusively, on how the changing pattern of cross-border linkages impacts financially-developed EMs with (partially) open capital accounts. For these countries, linkages with AEs and financial centers—the “core” nodes of the global network—dominate other channels. In Figure 4, these linkages are highlighted in the dotted-line box, with arrows representing the direction or the linkages from their origin. Of course, the web of cross-border linkages is infinitely thicker and more complex than this stylized representation, as there are many channels through which AEs, EMs and low-income countries (LICs) are interconnected. Some of these linkages are represented outside the dotted-line box in Figure 4. Among others, these could stem from reserve accumulation decisions in large EMs, or trade and commodity price linkages among EMs, AEs and LICs. Indeed, large EMs have been shown to have important macroeconomic spillovers for their regional LIC neighbors (IMF, 2011a). Similarly, rapid growth in large EMs has had a large impact on global commodity prices, shifting the terms of trade of other EMs and LICs. These linkages are undoubtedly important but are not analyzed here, in part because of data limitations (Box 2), but also because they are less likely to play a major role in the propagation of global shocks.

7. Relation to other staff work. This paper complements recent staff work on financial linkages presented in IMF 2009a, 2010b and 2010c. It also builds on recent staff work on cross-border capital flows (IMF 2010a and 2011e). One contribution of this paper is to bring together those aspects of the global financial infrastructure that are relevant for the design of a GFSN. Finally, this paper should be read in conjunction with the above-mentioned Systemic Crises paper, which focuses more explicitly on the triggers, propagation, and policy responses to past systemic crises, with a view to ascertaining whether the existing global financial safety net is adequate to deal with future systemic shocks.

Box 2.

Data Sources and Limitations

A number of data gaps preventing a full-fledged understanding of financial linkages were identified in previous staff work (IMF, 2010b) and efforts are ongoing to address some of these gaps, as discussed in a recent Board paper (IMF, 2011f). Understanding EM financial linkages is also hampered by the fact that bilateral data coverage for EMs is generally less complete than for AEs. Milesi-Ferretti and others, 2010, for instance, found that coverage for AEs is very satisfactory while it varies for EMs, being particularly low for Middle-Eastern oil exporters and ranging between 2/3 and 3/4 for a number of large EMs. Moreover, data is released with substantial lags. With these caveats, two main datasets were used throughout this paper:

Information on foreign bank claims was obtained from the BIS Consolidated Banking Statistics (immediate borrower basis). These statistics were collected on a group worldwide-consolidated basis, including the claims of subsidiaries and branches. However, only a subset of source countries (24) reported data consistently through the period 1999–2009. Moreover, only a few EM countries participated in recent years as source countries. One important caveat is that BIS consolidated banking statistics have a few breaks in the series. These breaks have not been taken into account in the analysis.
- Information on cross-border portfolio holdings was obtained from the IMF’s Coordinated Portfolio Investment Survey (CPIS), the annual survey of bilateral portfolio holdings. This data has some well-known limitations (Lane and Milesi-Ferretti, 2008, and Milesi-Ferretti and others 2010). First, not all the economies participate in the survey, including some that are likely substantial holders of external assets (these include some oil-exporting economies with large sovereign wealth funds, offshore centers, and economies with large holdings of official reserves or portfolio assets, such as China and Taiwan province of China). Second, there may be under-reporting of cross-border assets, including because of the incomplete institutional coverage of the survey. Third, the survey may not capture the portfolio holdings of entities resident in a given reporting country but owned by foreign investors. Similarly, holdings on residents in a financial center typically do not capture their ultimate destination. These limitations imply that a country’s implied external liabilities (as computed from the claims on the country held by residents in countries reporting cross-border claims) are typically below those reported in the country’s International Investment Position. To overcome some of these shortcomings, efforts to increase the frequency and to shorten the timeliness of the data and to collect data on the institutional sector of foreign debtors on an encouraged basis are ongoing. The implementation of these enhancements, beginning with the 2013 data, and efforts to increase the number of the participating countries, are also part of the G-20 Data Gaps Initiative.

This paper does not consider foreign direct investment (FDI), an investment class generally viewed as relatively stable and driven by longer-term considerations. One caveat is that the increased use of special purpose vehicles and other financial conduits by direct investors may suggest that not all FDI may be as stable as normally held. Finally, the analysis of banking sector linkages does not include off-balance sheets positions owing to data limitations, although these linkages are likely to be important for some countries.

III. Snapshot of Cross-Border Financial Linkages

8. A bird’s-eye view. The key stylized fact is that AEs still dominate cross-border financial linkages (Figure 5). In particular, more than 90 percent of claims issued by residents in EMs are held by residents in AEs or financial centers, while the share held in EMs is in most cases fairly small, generally 5 percent or less.⁵ This is true across asset classes and regions with the exception of debt holdings of Asian EMs for which linkages to EMs are relatively more important (Milesi-Ferretti and others, 2010). It is important to note that cross-EM linkages have increased very rapidly over the last decade (Table 1), but because of the low initial base they are still dwarfed by linkages emanating directly from AEs.

Table 1.

Cross-Border Portfolio Claims (US$ billion)

Source: CPIS; and IMF staff calculations

Table 1.

Cross-Border Portfolio Claims (US$ billion)

	Equity			Debt
	2001	2009	% change	2001	2009	% change
AE claims on AEs	4,357	9,732	123	4,875	1 5,813	224
AE claims on EMs	244	1,560	539	255	679	167
EM claims on AMs	43	269	530	35	187	430
EM claims on EM	1	41	3,668	5	29	534

Source: CPIS; and IMF staff calculations

Table 1.

Cross-Border Portfolio Claims (US$ billion)

	Equity			Debt
	2001	2009	% change	2001	2009	% change
AE claims on AEs	4,357	9,732	123	4,875	1 5,813	224
AE claims on EMs	244	1,560	539	255	679	167
EM claims on AMs	43	269	530	35	187	430
EM claims on EM	1	41	3,668	5	29	534

Source: CPIS; and IMF staff calculations

9. Core nodes. An essential feature of the global financial system is that it has relatively few countries at its “core” (Box 3). Moreover, the same countries tend to dominate the system across different asset classes. These “core” nodes, from both the source and recipient perspectives, are mostly AEs or financial centers; only a few EMs appear in this core list.⁶ This concentration of global financial activity within a handful of countries has been identified in previous research as a critical feature of the current global financial system (von Peter, 2007, and IMF, 2010c).

10. Overlapping nodes. Virtually all countries that are singled out as core in one asset class are core nodes for at least two of the three asset classes considered here, and many of them for the three classes. This overlap adds a further element of complexity to the global financial system.⁷ In addition, the large overlap between the top sources and top recipients suggests that shocks can be transmitted in both directions, significantly accelerating the spread of shocks. Indeed, the interplay among different asset classes has been singled out as a defining feature of the 2008 global crisis, when a shock originating in one specific corner of the U.S. financial system—subprime mortgages—was amplified through multiple linkages across asset classes and borders (Bordo and Langdon, 2010, and IMF, 2010c).

11. Have EMs been overlooked? Many large EMs do not participate in the databases considered here, and CPIS data do not provide a geographic breakdown of reserve assets by individual holder. However, using a more comprehensive database including reserve assets, Milesi-Ferretti and others, 2010, confirm that EMs still account for a small part of cross-border financial linkages. In 2007, the share of emerging Asia including China in external assets holdings was only about 5 percent of total global external assets; the same was true for external liabilities. In addition, China, which has supplied large savings to the U.S., still accounts for a limited share of U.S. asset market capitalization (Box 4).

Box 3.

What Countries Are “Core” Nodes?⁸

To gauge what countries are core to the global financial system across the three asset classes considered here (portfolio equity, portfolio debt, and bank claims), all source and recipient countries were ranked—for each asset class—according to their importance as sources and recipients on a bilateral basis. The top global source and borrower countries were then identified as those countries that recurred most frequently in these bilateral rankings (Box Figure). This approach to “core” nodes complements the measure of interconnectedness used in IMF (2010b), in two aspects:

It took into account both the asset side (i.e., top sources) and the liability side (i.e., top recipients), while IMF (2010b) focused only on the latter.
It identified core nodes across three asset classes while IMF (2010b) covered bank claims only.

This exercise revealed some interesting, albeit unsurprising, features of the global financial system (Annex III presents the complete rankings). First, most core nodes were AEs or financial centers; the only EMs classified as core nodes were Brazil for top sources, and Brazil and Korea for top recipients. Second, virtually all countries that were singled out were core nodes for at least two asset classes, and many of them for the three classes (Germany, France, Japan, Ireland, Netherlands, U.K., and U.S. appeared as core global sources and recipients for all three classes).

Box 4.

China’s Supply of Savings and Holdings of U.S. Financial Assets⁹

While AEs still hold the bulk of the stock of cross-border claims, on a flow basis, some EMs have been increasingly large providers of net savings to advanced countries (Box Figure). China, in particular, has been a large supplier of net savings to AEs, especially the U.S. For example, during 2001–2010, the U.S. imported US$5.8 trillion from the rest of the world (measured as the cumulative current account balance over the period). This flow of savings was largely supplied by a limited number of economies, most notably Japan, China and oil producers and took largely the form of accumulation of official reserves.

The role of China as a source in the network of cross-border linkages is difficult to establish, though, given that China does not report data to the BIS and CPIS databases. The U.S. Treasury’s Treasury International Capital System (TIC) data, however, allows taking stock of China’s holdings of a core AE (Box Figure). This data confirms the stylized fact that, when looking at stocks, the penetration of China—and EMs more generally—in this core node remains limited and concentrated in sovereign debt markets.

12. Bank ownership linkages. Cross-border bank ownership has provided an important impetus to cross-border financial linkages between AEs and EMs. This is clear in a novel dataset of cross-border bank ownership put together by staff. This shows that the importance of cross-border banking groups has grown over time, most notably in European and Latin American EMs, where on average assets belonging to foreign-owned groups represent between 30 and 40 percent of total domestic assets—with the group parents mostly residing in AEs and financial centers (Figure 6). By contrast, the importance of cross-border asset ownership is much lower in AEs. Moreover, cross-border groups are still largely owned by a parent bank residing in AEs and financial centers.

13. Takeaways. (i) Cross-border financial linkages are still overwhelmingly to AEs; and (ii) there are relatively few countries, mostly AEs and financial centers, that act as “core” nodes in the global financial system.

IV. Cross-Border Financial Linkages and Shock Transmission

14. Capital flows and shock transmission. Shifts in cross-border exposures, especially in the form of rapid synchronized deleveraging, are a key source of systemic risks.¹⁰ EMs are most exposed to such systemic shifts given that the bulk of their exposures are toward core nodes (AEs or financial centers), which tend to be propagators of global shocks. Recent staff work on capital flows to EMs has highlighted a few important stylized facts about capital flows to EMs. While the lessons from this analysis can be generalized to all countries, many features are particularly true for EMs:

Deleveraging. As evidenced during the crisis, many EMs experienced a sharp reversal in net capital inflows (Figure 7), which was in many cases driven by a change in external liabilities (that is, a sudden stop in capital inflows), with relatively little action on external assets (limited capital outflows). This is consistent with a more general pattern, whereby changes in EMs’ gross external liabilities have often been associated with changes in net external liabilities, far more than it is the case in AEs.
Synchronization. Shifts in cross-border exposures can be highly synchronized, especially at times of stress. Episodes of capital inflow surges normally start at different times, likely a reflection of country-specific circumstances and pull factors,¹¹ but often end together within a narrow time period (Figure 9), as seen for example during the sudden stop episodes of 1997–98 and 2008–09. This suggests that behind these reversals are exogenous factors, such as sudden shocks to global risk appetite. This feature also explains why an insurance mechanism against sudden shifts in cross-border exposures driven by aggregate or global shocks cannot be based exclusively on local or regional risk-sharing mechanisms (Holmström and Tirole, 1998, and Levy-Yeyati, 2010). The evidence discussed in Section V that there remain strong geographical patterns in cross-border networks further reinforces the case for a global insurance mechanism.
Volatility In the post-crisis recovery there has been a shift toward flows that are historically relatively more volatile than others, portfolio flows (Figure 10 and IMF, 2011e). Moreover, compared to past episodes of capital flow surges, the average pace of portfolio inflows during this ongoing wave has more than quadrupled. This may increase the risk of sudden reversals. As noted below, portfolio flows are also the flows that appear to be more closely related to global factors—and as a result potentially more exposed to global shocks.
Global drivers. Given the evidence above on the synchronization of sudden stops and the fact that most cross-border claims on EMs are held by AEs, it is perhaps surprising that empirically the role of global factors in driving capital flows is at times found to be relatively small (IMF, 2011b). Indeed, the empirical analysis discussed in Annex IV suggests that global factors could explain only around 25 percent of variations in total gross inflows to EMs. The importance of global factors for portfolio inflows, however, is likely to be much higher—the estimates in Annex IV put it, on average, at around 50 percent of total variation in portfolio inflows to EMs.¹² Moreover, the importance of global factors tends to shift over time. In particular, capital flows to EMs tend to be large during periods of low global interest rates, low global risk aversion, and high growth differentials between emerging markets and advanced economies (Figure 11). These shifts over time may explain why previous empirical research has not produced a consensus view on the relative importance of push (external) vs. pull (domestic) factors (Box 5).

This pattern can be seen in the time series correlation between gross and net capital inflows (Figure 8). This correlation is often close to zero in non-EMs, while it tends to be closer to one in EMs, implying that shifts in gross inflows are normally offset by changes in outflows in AEs but not in EMs. One possible explanation of this pattern is that because AEs are more interconnected both in terms of inward and outward linkages, they have more “degrees of freedom” in offsetting one change in one linkage with another adjustment elsewhere in the network—this is an illustration of the general principle that more links can improve risk diversification. On the other hand, EMs—generally less interconnected—lack this flexibility and are as a result subject to the higher one-way risk of deleveraging. It follows that EMs stand to benefit the most from a global insurance mechanism against this risk.

15. Accounting for shifts in exposures. The stylized facts just summarized show that shifts in cross-border exposures can be highly synchronized and volatile. What are the factors that help explain these features? Global drivers—whose impact can be assessed both in terms of quantities and prices—are important in explaining these shifts, while shallow capital markets in recipient country can amplify the shocks. Moreover, there is some evidence of shock transmission through “imitation”—that is, countries with common characteristics being hit harder. Over a longer horizon, structural shift in global asset allocation is also another driver of shift in cross-border exposures and shock transmission.

Box 5.

The Importance of Global Factors for Capital Inflows to EMs: Literature Review¹³

There seems to be lack of consensus in the empirical literature on the quantitative importance of global factors as drivers of capital flows to EMs. This box briefly summarizes some findings at the opposite extremes of this literature.

The first generation of empirical literature, inspired by the surge of capital flows into EMs in the 1990s, favored the push view. Summarizing the early literature, Fernandez-Arias and Montiel, 1996, concluded that falling U.S. interest rates played a dominant role in driving capital flows to developing countries. Fernandez-Arias, 1996, further estimated that the fall in international interest rates explained 86 percent of the increase in portfolio flows in 13 middle income countries between 1989 and 1994. Chuhan and others, 1993, found that global factors such as the fall in U.S. interest rates and the slowdown in the U.S. economy explained about half of the increase in equity and bond flows to nine Latin American countries. For Asian countries, they estimated that external factors accounted for about one-third of portfolio flows into the region.

More recent literature, however, has suggested that pull factors and country fundamentals are relatively more important. Using a variance decomposition approach, Mody and others, 2001, concluded that domestic pull factors dominated push factors in explaining a large portion of the forecast variance. More recently, IMF, 2011b, showed that global factors explained 20 percent of the variation in net capital flows into EMs.

Despite these differences in view about the importance of global factors for total capital flows into EMs, there has been less divergence in the literature that push factors have played a more significant role in explaining certain type of flows, e.g., portfolio bond flows. Summarizing two strands of research carried out at the Bank of England, Ferrucci and others, 2004, noted that push factors, and in particular U.S. short-term interest rates, explained two thirds of the compression in EM bond spreads. The contribution of push factors was found to be less significant for banking flows than for other asset classes but almost as important as pull factors.

The lack of a consensus view on the relative importance of push and pull factors may simply reflect the fact that their respective roles vary substantially over time and across countries, a point echoed by Lane, 2009.

Asset price co-movements. Consistent with the evidence of increased global financial linkages, it is not surprising that correlations among a broadening range of asset markets have been increasing over time¹⁴ (Figure 12), with common factors explaining a significant fraction of EM cross-country asset price variation (Fernandéz-Arias and Levy-Yeyati, 2010). Indeed, common factors have become more important over time in driving EM external yields—for instance, the contribution of the first principal component to the total variation in EM external yields, already high in the first few years of the past decade, has grown further in more recent years, reaching almost 80 percent (Figure 13). Moreover, the decomposition of the total EM spillover index shown in Figure 1 indicates EMs have been affected mostly on the receiving end (Figure 14), meaning that EMs tend to be on average net receivers of global/AM shocks (Annex V discusses the methodology employed here). Finally, asset prices also show clearly how financial linkages shift over time. These shifts can be especially abrupt at times of crises—the 2008 global crisis is again a case in point—highlighting how the transmission of financial shocks can be highly non-linear. In this regard, global risk aversion (captured here by the VIX index) has become an increasingly important source of volatility for global markets, and EMs in particular, with a spike at the time of the global crisis.¹⁵
Common characteristics. The global crisis has also shown that shock could be transmitted through investors’ perception of countries’ common characteristics. Even countries that are not considered “systemic” ex ante in terms of economic size and financial/trade linkages could become the epicenter of a systemic event, as their crisis could serve as a “wake up” call to creditors in core nodes, causing them to pull back from other countries sharing characteristics similar to those at the epicenter of the crisis. Latvia and other Eastern European countries in the 2008 global crisis illustrate this point (Box 6).
Relative size of capital markets. Shocks to EMs are amplified by the relatively small size of their capital markets. The fact that cross-border linkages are still largely dominated by claims held in a few AEs and financial centers has a counterpart in the absolute and relative size of domestic capital markets (Figure 15). Whereas EMs represented about a third of world GDP in 2009, their stock markets and bank assets were around one fifth of these asset classes on a global level; debt markets were an even smaller share, less than one tenth when public and private debt markets are combined. Hence, even a small shift in portfolio allocations from AEs to EMs could easily overwhelm EMs’ absorptive capacity. For example, a reallocation of 1 percent of assets from AE markets stock, public debt or bank assets corresponds to a shift of between 4 and 6 percent in terms of EM market size, and even more (20 percent) for private debt markets. Given that the larger financial markets in AEs are also generally deeper and more liquid than their counterparts in EMs, the impact of such shifts may be even larger than suggested by differences in size alone.
Shift in EM assets. Given rapid growth of EM holdings of external assets, some large EMs’ shift in their asset allocation could eventually have significant repercussion to the global financial markets. This is particularly true for China (Box 7).

Box 6.

Linkages and Shock Transmission by Association: The Eastern European Experience¹⁶

This paper focuses mostly on cross-border direct financial linkages. Put simply, two countries are linked if they trade assets directly. This box explores a broader concept of linkages emerging from a combination of actual financial linkages and similarity in policy frameworks and fundamentals.

Similarity in policy frameworks and fundamentals were very relevant in Eastern Europe during the 2008 global crisis. For example, Bulgaria, Estonia, Latvia, and Lithuania shared two critical features at the onset of the crisis: a significant presence of Western European banks and a hard peg to the euro (currency boards).¹⁷ These countries also shared common vulnerabilities—rapid credit growth, asset price inflation, and large current account deficits. As a result, despite weak direct and indirect financial linkages between Bulgaria and the Baltics, a large shock to Latvia was seen as potentially disruptive for Bulgaria and other Eastern European countries.

Actual financial linkages obviously explained part of the perception that Eastern European currency boards could come under stress. For example, Estonia, Lithuania, and Latvia all borrowed significant amounts from Swedish banks (Box Figure). The common creditor argument to some extent was at work in the Baltics: localized stress in one of the Baltic countries could have weakened the Swedish banking system and have led Swedish banks to deleverage in the other Baltic countries. However, the direct links between Swedish banks and Bulgaria’s lenders, mostly Greece, Italy, and Austria, (Box Figure) were weak: Sweden’s claims on Greece, Italy and Austria together only accounted for about 3 percent of Sweden’s foreign bank claims in 2007.

Beyond direct links, common policy frameworks and vulnerabilities played a role in associating the fate of Bulgaria and the Baltics during the crisis. Intense distress in one country with a currency board and large liabilities to foreign banks could act as a “wake-up” call for investors and domestic depositors who could conclude that other similarly-situated countries may face similar distress. If this belief were to become entrenched, Bulgaria and the Baltics would be closely interconnected even if the (indirect) links between Sweden, on one hand, and Greece, Italy, and Austria, on the other hand, are not very significant.

Box 7.

Long Term Spillovers of China’s Portfolio Allocation Shifts¹⁸

China is projected to contribute to more than one third of global net wealth accumulation between 2010 and 2015, due to continued rapid economic growth and high savings (net wealth accumulation is here defined as net investment plus increase in net foreign assets). The allocation of China’s vast new wealth will have increasingly important implications for both the domestic and global financial markets.

China’s non-reserve asset holdings show high degree of home bias, a result of capital controls and an exchange rate that is substantially below the level that is consistent with medium-term fundamentals, among other factors. If the current configuration of asset holdings (as measured by their portfolio shares) were to persist in the medium term, the demand for Chinese assets would outpace the supply of Chinese assets—whose net supply grows with net investment in the economy. Simple market clearing conditions would imply significant increases in China’s asset prices and downward pressures in some offshore markets (Box Figure 1)

Policy responses could alter the nature of these spillovers. China’s central bank could reduce asset price inflation by accumulating more reserves through the sterilized purchase of foreign assets. Staff estimates that further reserve accumulation by China in the order of US$600 billion during 2011–2015 would be needed in addition to the US$2 trillion accumulation under the baseline scenario to keep the real price of Chinese financial assets in line with the real price of U.S. assets. The spillover on third markets would be significant, however, especially if China were to increase its percentage holdings of assets from these third markets relative to the current portfolio allocation (Box Figure 2).

16. Macroeconomic effects of financial linkages. Financial linkages explain a large portion of the macroeconomic spillovers from AEs to EMs. To gauge the macroeconomic importance of cross-border financial shocks, four pieces of empirical evidence were considered:

First, a vector autoregressive model was estimated on two economic blocs, AEs and EMs. Impulse response functions from this model show that shocks from the AE bloc have a quantitatively important effect on the EM bloc, with the reverse not being the case (Figure 16). This exercise also confirms that an important fraction of growth spillovers can be attributed to financial shocks in the AE bloc.
Second, an empirical model of cross-border bank groups’ behavior Regression analysis on bank groups illustrated the mechanics of shock transmission within groups, and hence across borders (Figure 17). The empirical evidence discussed in greater detail in Annex VI show that these groups can be important conduits for cross-border macroeconomic shocks. More specifically, the lending behavior of EM subsidiaries is affected by their own leverage and liquidity conditions as well as by the local macroeconomic conditions. However, shocks to the parent’s financial conditions (liquidity, capital adequacy, profitability or non-performing loans, as well as macroeconomic conditions in the parent’s home country) are also important determinants of the subsidiary’s lending behavior in EMs. By contrast, parent-to-subsidiary effects are less statistically robust when a subsidiary is located in an AE. This reinforces once again the point that EMs tend to be particularly subject to shocks emanating from AEs.¹⁹
Third, pre-crisis interconnectedness and country growth performance in the global crisis. A test of the role of interconnectedness during the 2008 global crisis was conducted using a variation of regression analysis employed in previous staff work (IMF, 2010e, Table 2). This approach consisted in relating output contraction during the 2008 global crisis to country fundamentals, including external debt, growth in trading partners, international reserve coverage of short-term gross financing needs, and an index of external vulnerability. The variation considered for this paper was to replace the latter with the network indices of interconnectedness and concentration—in-degree and HHI, respectively (Annex II discusses the construction of these indices). Once controlling for other fundamentals, more diversified countries (as measured by higher in-degree) suffered less output contraction than less diversified countries (Table 2).²⁰ This finding is thus consistent with the risk-diversification argument that higher interconnectedness can help smooth shocks. However, countries with more concentrated exposure (as measured by a higher HHI) suffered a more pronounced output contraction,²¹ again consistent with the theoretical prediction that high concentration can be a shock amplifier.
Fourth, a total spillover index based on real variables. Against the clear trend of increased financial interconnectedness, it is intriguing that, at least over the last decade, there has been much less clear-cut evidence of an upward trend in interconnectedness between real variables in different regions—Figure 18 shows the analogue of the total spillover index on financial variables shown in Figure 1 for industrial production indices in G7 countries, a group of EMs, China, and the VIX index. This “real” interconnectedness index remained broadly stable in the years leading up to the crisis but jumped with the crisis, and has remained elevated since then.²² Is there a puzzle from the lack of trend in real interconnectedness before the crisis, given that financial interconnectedness was instead on an increasing trend? Insights from network theory offer a (tentative) explanation of this phenomenon. During the period of the “great moderation”, shocks could be dispersed across the network because they were relatively small and did not hit core nodes. As a result, interconnectedness among real variables did not display a trend, even though the underlying plumbing of the financial system was becoming increasingly interconnected. This last feature, however, increased the system’s fragility to a systemic shock, which—when it hit—caused great disruption across the system and a spike in measured real spillovers.

Table 2.

Regressions for percentage change in real output between peak and trough for EMs during the global crisis

Source: Database from “How Did Emerging Markets Cope in the Crisis” (IMF 2010), BIS and IMF CPIS.

17. Takeaways. (i) As financial linkages mainly emanate from AEs, shifts in cross-border exposures are a key channel for the propagation of shocks globally; EMs, in particular, are exposed to such shifts, which can be extremely large, given EMs relatively shallow financial markets; (ii) growing global financial linkages also imply increasing importance of global drivers for EM asset prices movements; (iii) both macro and micro (banking) empirical results show that cross-border financial shocks from AEs and network interconnectedness could have significant macroeconomic impact on EMs; and (iv) empirical evidence also suggests that, conditional on a systemic shock occurring, more interconnected countries suffered a smaller output decline during the crisis; however, countries with more concentrated exposure suffered a more pronounced output contraction.

V. What Explains Cross-Border Linkages?

18. Motivation. This section seeks to explain the strength and drivers of cross-border linkages. By relating the size of cross-border financial linkages to structural and economic features of source and recipient countries, this analysis highlights the determinants of channels through which shocks can potentially be transmitted. Thus, this analysis may help distill some general principles that could underlie a GFSN.

19. Does the size of linkages help predict potential shocks? To further motivate this section, a simple exercise was carried out whereby the change in bilateral exposure between 2007 and 2008—that is, during the global crisis—was related to a number of underlying determinants, including the size of the initial exposure (Table 3). The key finding was that the size of the initial linkage helped explain the size of deleveraging during the global crisis: the larger the initial exposure, the larger the subsequent deleveraging.²³ Moreover, EMs experienced larger deleveraging than AEs. These results thus motivate a search for determinants of cross-border linkages, as stronger linkages can in principle lead to deeper deleveraging once a shock hits.

Table 3.

Deleveraging regression

Dependent variable: Change in log claims between 2007 and 2008 ¹,^2/

Source: BIS, CPIS, and staff estimates.

^{^1/}Bold denotes significant at 1, 5 or 10 percent level. Other control variables include bilateral exports, financial openness, financial depth, exchange rate depreciation and financial center dummies.

^{^2/}Change in log claims are not corrected by valuation change which could be significant in the case of portfolio equity.

Table 3.

Deleveraging regression

Dependent variable: Change in log claims between 2007 and 2008 ¹,^2/

	Portfolio debt		Portfolio equity		BIS bank claims
	Coeff	P> \|t\|	Coeff	P> \|t\|	Coeff	P> \|t\|
Log of claims in 2007	-0.09	0.02	-0.16	0.01	-0.08	0.01
EM recipient dummy	-0.23	0.15	-0.17	0.07	-0.13	0.07
EM source dummy	0.20	0.12	0.26	0.07	-0.19	0.09
Observations	1,262		2,576		1,212
R-square	0.11		0.17		0.08

Source: BIS, CPIS, and staff estimates.

^{^2/}Change in log claims are not corrected by valuation change which could be significant in the case of portfolio equity.

Table 3.

Deleveraging regression

Dependent variable: Change in log claims between 2007 and 2008 ¹,^2/

	Portfolio debt		Portfolio equity		BIS bank claims
	Coeff	P> \|t\|	Coeff	P> \|t\|	Coeff	P> \|t\|
Log of claims in 2007	-0.09	0.02	-0.16	0.01	-0.08	0.01
EM recipient dummy	-0.23	0.15	-0.17	0.07	-0.13	0.07
EM source dummy	0.20	0.12	0.26	0.07	-0.19	0.09
Observations	1,262		2,576		1,212
R-square	0.11		0.17		0.08

Source: BIS, CPIS, and staff estimates.

^{^2/}Change in log claims are not corrected by valuation change which could be significant in the case of portfolio equity.

20. Methodology and data. To explain cross-border linkages, regression analysis was used to relate a country’s cross-border portfolio and bank liabilities to structural and economic features of source and recipient countries (Annex VII provides the details of this exercise). Consistent with previous literature, the empirical model was based on a “gravity” equation of financial linkages between a source and a destination country, where both source and destination groups included EM and AE countries.²⁴ A number of time-varying and time-invariant determinants were considered as explanatory variables of the bilateral linkages: geographical and historical “gravity” variables (distance, time difference, common legal system, and common language); economic development (size, income per capita, goods trade), financial developments (financial depth and capital account openness); and regional group dummies. Year fixed effects were also added to control for global trends. The model was estimated for both the 2009 cross-section of bilateral exposures and the panel data from 2001–09 using both country-pair random effects, and source and recipient country fixed effects.

21. Key findings. Findings were generally in line with the existing literature and broadly consistent across asset classes. More specifically, the key findings can be summarized as follows (Figure 19 and Annex VII):

Geography and history. Geography matters: distance between source and recipient, generally seen as a proxy for information quality, is a statistically significant determinant of cross-border financial linkages, with higher exposure being built up toward countries that are nearer. The effect is present for all asset classes, and is particularly strong for bank claims. Historical and cultural factors also have a role in explaining the strength of linkages: countries with the same legal system and common language tend to have stronger exposure, although there is some differentiation across asset classes in terms of the statistical significance and strength of these factors. Countries that share common language and legal system tend to have stronger cross-border exposures, especially for equity and bank claims.
Economic factors. Economic size of both source and recipient matters as a determinant of cross-border exposure. Larger countries—both source and recipient ones—have larger cross-border exposures. More developed countries (as measured by overall GDP and per capita GDP, respectively) tend to have larger portfolio exposures. Combined, size and income level explain about 40 percent of the interquartile differences across all asset classes.²⁵ Not surprisingly, increased trade integration is also associated with higher cross-border exposures. Finally, a common currency tends to be associated with stronger linkages.
Financial factors. Across asset classes, increased financial development in both source and recipient countries feeds stronger cross-border financial linkages—this effect is strongest for bank claims. In a similar fashion, source and recipient countries that are classified as financial centers also develop stronger cross-border exposures. For source countries, this feature—already identified in previous research, e.g. IMF, 2010c—is stronger for portfolio debt and bank claims exposure. Finally, recipient countries with more open capital accounts tend to have stronger portfolio exposure across all asset classes—the effect of capital account openness is not statistically significant for source countries.
Regional groups. Country group dummies for both source and recipient sides help capture differences in the level of exposures across groups (Table 4). From the recipient side, G-7 countries remain dominant. This result is statistically significant for portfolio debt and equity, but not so for bank claims. EMs in general have lower exposures compared to advanced economies. G-7 dominance as source countries especially strong across asset classes.

Table 4.

Coefficients for country group dummies ^1/

Source: BIS, CPIS and staff estimates.

^{^1/}Comparison group is other advanced countries group. Bold denotes significance at 1, 5 or 10 percent level.

^{^2/}BIS data do not include EM source from Asia or the Middle East.

Table 4.

Coefficients for country group dummies ^1/

Recipient	Equity	Debt	Bank claims ^2/
G-7	0.57	0.68	0.21
European EMs	-0.49	-0.60	-0.46
Middle east EMs	-0.44	0.078	-0.96
Asian EMs	0.08	-0.61	-0.11
Latin American EMs	-0.38	-0.13	0.31
Source
G-7	1.53	1.45	1.25
European EMs	-1.36	-2.1	-1.48
Middle east EMs	-0.63	-1.1
Asian EMs	-0.89	-0.41
Latin American EMs	-0.92	-1.2	-2.23

Source: BIS, CPIS and staff estimates.

^{^1/}Comparison group is other advanced countries group. Bold denotes significance at 1, 5 or 10 percent level.

^{^2/}BIS data do not include EM source from Asia or the Middle East.

Table 4.

Coefficients for country group dummies ^1/

Recipient	Equity	Debt	Bank claims ^2/
G-7	0.57	0.68	0.21
European EMs	-0.49	-0.60	-0.46
Middle east EMs	-0.44	0.078	-0.96
Asian EMs	0.08	-0.61	-0.11
Latin American EMs	-0.38	-0.13	0.31
Source
G-7	1.53	1.45	1.25
European EMs	-1.36	-2.1	-1.48
Middle east EMs	-0.63	-1.1
Asian EMs	-0.89	-0.41
Latin American EMs	-0.92	-1.2	-2.23

Source: BIS, CPIS and staff estimates.

^{^1/}Comparison group is other advanced countries group. Bold denotes significance at 1, 5 or 10 percent level.

^{^2/}BIS data do not include EM source from Asia or the Middle East.

22 Interconnectedness of bank groups and geographical distance More evidence on the evolution of interconnectedness is derived from the dataset on bank groups introduced earlier. Starting with the concentration of assets associated with cross-border banking operations, there has not been a common trend across source countries, although in some notable cases there has been a marked increase in the asset concentration (as for banks with parents from Spain or Hong Kong SAR) of their external operations²⁶ (Figure 20). However, there has been a much more consistent trend toward a lower average “distance” between subsidiaries and their parent. In other words, cross-border groups have tended to move “closer to home,” emphasizing the role of regional agglomeration, similar to the above regression findings.

23. Takeaways. This section provides some insights on features that could be taken into account in the design of a GFSN. (i) Larger and more developed countries tend to have stronger cross-border linkages. This is reinforced by financial development, both in the recipient and source economies. As a result, the design of a GFSN needs to be mindful of the importance of financially-developed economies, both as sources of shocks as well as possible targets. (ii) Geography matters. Notwithstanding the globalization of financial markets, there are still forces pushing toward geographical “agglomeration”, namely a tendency for linkages to be stronger with closer countries. This could limit the benefit of risk-sharing scheme at a regional level, and argues for a global response; and (iii) The role of geographical factors overlaps with that of historical factors, such as a shared language or legal system, which tend to create stronger cross-border exposures. This suggests the existence of “special” linkages, explaining, for example, the large relevance of Spanish banks in Latin America.

VI. Conclusions

24. This paper shows that cross-border financial linkages have increased dramatically over time, both in terms of growing cross holdings of external assets and co-movements of asset prices across asset classes and economies. These linkages are still dominated by a few AEs and financial centers, which form “core nodes” in the system. EMs’ cross-border holdings, despite having experienced rapid growth in the last decade, remain relatively small, while the bulk of EMs’ liabilities are held by core nodes. EMs, with shallow domestic capital markets, are particularly exposed to large shifts in cross-border exposures. The empirical evidence in this paper suggests that shocks can be highly synchronized across global financial markets and their transmission non-linear, potentially carrying significant real costs to EM economies.

25. Efforts to increase the resilience of the global financial system face a delicate trade-off between the benefits at the country level from increased international risk diversification—a force that pushes toward increasing the system’s overall interconnectedness—and the increased systemic risk that this heightened interconnectedness generates at the global level. National defenses, including via self-insurance, may not be able to eliminate the externalities intrinsic in an interconnected financial network. To the extent that national defenses are insufficient to reduce the source of volatility and instability inherent with the global financial system, regional and, especially, global financing mechanisms have a role to play in cushioning the impact of residual volatility on countries and the system at large. The recent global crisis underscores the value of an effective global mechanism to coordinate liquidity injections and other policy responses: in 2008–09, a number of countries with relatively strong fundamentals (“crisis bystanders”) were hit later and less severely than countries with weaker fundamentals, and were able to recover from the crisis more rapidly; but they still suffered a deep output contraction. This, combined with the limited opportunities to diversify against the risk of cross-border shocks, points to the significant gains from a global financial safety net that is effective in forestalling localized liquidity runs to turn into systemic crises.

Annex I. Country Groupings

Table I.1 lists the country groupings used in this paper, subject to data availability. Following Lane and Milesi-Ferretti, 2008, financial centers are identified in this paper as those economies with very large external assets and liabilities relative to GDP and a large financial system specializing in the production of intermediary financial services.

Table I.1.

Countries Included in the Sample

^{^1/}Exclude Lebanon and Panama, which are financial centers.

^{^2/}Excludes the UK, which is a financial center.

Table I.1.

Countries Included in the Sample

Emerging Markets ^1/
Asia	Europe	Middle-East & Africa	Latin America
China	Albania	Egypt	Argentina
India	Belarus	Georgia	Bolivia
Indonesia	Bosnia & Herz.	Jordan	Brazil
Korea	Bulgaria	Kazakhstan	Chile
Malaysia	Croatia	Morocco	Colombia
Mongolia	Estonia	Pakistan	Costa Rica
Philippines	Hungary	South Africa	Dominican Rep
Sri Lanka	Israel	Tunisia	Ecuador
Thailand	Latvia		El Salvador
Vietnam	Lithuania		Guatemala
	Macedonia		Jamaica
	Montenegro		Mexico
	Poland		Paraguay
	Romania		Peru
	Russia		Uruguay
	Serbia
	Turkey
	Ukraine
G-7 excl UK ^2/		Other advanced Economies
		Oceania	Europe
Canada		Australia	Cyprus
France		New Zealand	Denmark
Germany			Finland
Italy			Greece
Japan			Iceland
US			Norway
			Portugal
			Spain
			Sweden
Financial centers
Asia	Europe	Middle-East & Africa	Latin America
Hong Kong, SAR	Belgium	Lebanon	Panama
Singapore	Ireland
	Luxembourg
	Malta
	Netherland
	Switzerland
	UK

^{^1/}Exclude Lebanon and Panama, which are financial centers.

^{^2/}Excludes the UK, which is a financial center.

Table I.1.

Countries Included in the Sample

Emerging Markets ^1/
Asia	Europe	Middle-East & Africa	Latin America
China	Albania	Egypt	Argentina
India	Belarus	Georgia	Bolivia
Indonesia	Bosnia & Herz.	Jordan	Brazil
Korea	Bulgaria	Kazakhstan	Chile
Malaysia	Croatia	Morocco	Colombia
Mongolia	Estonia	Pakistan	Costa Rica
Philippines	Hungary	South Africa	Dominican Rep
Sri Lanka	Israel	Tunisia	Ecuador
Thailand	Latvia		El Salvador
Vietnam	Lithuania		Guatemala
	Macedonia		Jamaica
	Montenegro		Mexico
	Poland		Paraguay
	Romania		Peru
	Russia		Uruguay
	Serbia
	Turkey
	Ukraine
G-7 excl UK ^2/		Other advanced Economies
		Oceania	Europe
Canada		Australia	Cyprus
France		New Zealand	Denmark
Germany			Finland
Italy			Greece
Japan			Iceland
US			Norway
			Portugal
			Spain
			Sweden
Financial centers
Asia	Europe	Middle-East & Africa	Latin America
Hong Kong, SAR	Belgium	Lebanon	Panama
Singapore	Ireland
	Luxembourg
	Malta
	Netherland
	Switzerland
	UK

^{^1/}Exclude Lebanon and Panama, which are financial centers.

^{^2/}Excludes the UK, which is a financial center.

Annex II. Network Analysis²⁷

This annex first provides a short review of network theory and then describes the methodology and data used in measuring a country’s financial interconnectedness with the rest of the world and the concentration of its cross-border exposures.

Insights from network theory

Network analysis. Many of the financial market phenomena described in this paper can be seen—using insights from network theory—as the result of increased interconnectedness in a financial network. Besides the staff work mentioned in the paper, a growing literature has examined the properties of financial networks and their implications for financial stability (see Haldane, 2009, for an excellent overview of the field; there have also been a number of application of network analysis to financial stability issues by staff, see, e.g., IMF, 2009a). In particular, the seminal contribution by Allen and Gale, 2005, showed that—in the absence of financial frictions—more interconnectedness increases risk diversification and makes the network more robust. Subsequent research has shown, however, that in the presence of financial frictions, high levels of interconnectedness may increase systemic risk (Battistion and others, 2009, and Gai and Kapadia, 2008). Finally, a number of papers have used a range of network statistics to describe the financial network and patterns of stress transmission (Garrat and others, 2011, Kubelec and Sa, 2010, von Peter, 2007).

Interconnectedness as a double-edged sword. The key insight of network theory applied to financial stability analysis is that interconnectedness is a double-edged sword. Interconnectedness has the potential of making a network more robust, but it also increases the risk of rare but devastating events. At low levels, an increase in interconnectedness improves the resilience of a network; at high levels, however, it may raise the network’s latent fragility and vulnerability to systemic breakdowns. The latent fragility of complex networks is a key rationale for the existence and appropriate design of a GFSN.

“Robust” networks. Recipient countries with a larger number of cross-border links tend to benefit from better risk diversification of sources,²⁸ as a localized shock hitting a single source country and the ensuing deleveraging can be more manageable. The opposite is true in a network with few links.²⁹ This diversification argument—formalized by Allen and Gale, 2005—means that interconnectedness can improve risk sharing and make the network more robust.

“Fragile” networks. A higher degree of interconnectedness can also increase the latent fragility of a financial network. If a highly interconnected source country (a core node) is hit by a large shock, the network might display a tipping-point property, as the shock can be propagated widely via the country’s large number of links to the rest of the network. This fragility results from a negative network externality: when an “average” country increases its connections to the network, it does not internalize the risk of increasing the network’s fragility to systemic shocks.

Nonlinearities: Asymmetric information and herd behavior. Besides allowing for wider shock propagation, more interconnectedness is also likely to result in incomplete information of the overall financial system. In this context, pure contagion and herd behavior could propagate shocks beyond direct trade and financial linkages. In the presence of incomplete information, financial integration strengthens investors’ incentives for herding behavior (Calvo and Mendoza, 1997). Moreover, unusual or unexpected events can trigger the perception of ‗immeasurable risk’, leading to a flight to quality (Caballero, 2009, interprets the global crisis through this lens).³⁰ Supporting this point, Caballero and Simsek, 2011, have shown that incomplete information in a complex financial network creates an environment prone to fire sales and liquidity crunches. In this environment, even countries that are not considered “systemic” ex ante could be the epicenter of a systemic event, as their crises serve as a “wake up” call to creditors, who could shed assets of other countries sharing similar characteristics. As argued in the Systemic Crises paper, it is thus important to go beyond the usual approach of focusing on “systemic” economies to assess systemic risks, as this approach tends to overlook “systemic” stress arising from smaller countries that may not be highly interconnected; ex post, such negligence could become costly.

Concentration risk. Additional risks arise when recipient countries have an unusually large concentration of exposures to a few sources. In these circumstances, a large shock hitting a main source could create more severe deleveraging than in an alternative situation where all links were uniform across sources. The importance of this risk has been documented empirically. For instance, during the 2008 global crisis, countries with the largest portfolio holdings in emerging Europe at the end of 2007 experienced the largest portfolio adjustment in 2008 (Galstyan and Lane, 2010).

Interconnectedness versus concentration. Interconnectedness and concentration are compatible concepts. As depicted in a stylized fashion in Figure II.1, the two countries have the same number of links and hence the same degree of interconnectedness with the network (based on the in-degree index introduced below). However, while the country on the left has uniform dependence across its sources—and therefore low concentration risk—the country on the right depends much more heavily on a single source country (no. 5), raising the risk that a shock in this source country will force the recipient country to deleverage rapidly.

Measuring interconnectedness and concentration

Empirical measures. Several measures have been developed in the literature to capture different aspects of network interconnectedness (see, for example, the overview in von Peter, 2007). One of these measures, “in-degree”, is useful to capture the notion of diversification in a network.³¹ For a recipient, this index is constructed as the share of “active” links to its sources to the total number of potential links. Here, a link is considered active if the stock of the recipient’s cross-border liabilities is above 0.2 percent of the recipient’s GDP. By normalizing a link to the recipient’s GDP, this approach emphasizes the importance of a link from the perspective of a recipient rather than a source, as frequently done in the literature.³² Holding constant the total stock of cross-border liabilities, a higher in-degree index simply means that a recipient has on average smaller exposures to its sources, abstracting from the cross-source distribution of claims. This distribution is instead at the core of measuring concentration risk via a (normalized) Hefindahl-Hirschman index (HHI). This type of index is commonly used as a measure of concentration, for instance in bank loan portfolios or among firms in an industry. This index is normalized to range between zero (no concentration) and one (only one source per recipient).

Specifically, in-degree and concentration are defined as follows:

In-degree is the number of locations a country borrows from. In order to come up with an economically meaningful measure, links smaller than 0.2 percent of the recipient’s GDP were disregarded. ³³ As the BIS and CPIS datasets cover different numbers of sources, in-degree was normalized by the number of sources so that it always lies in the interval between 0 and 1, making it comparable across different types of claims and datasets:
$I n - \deg r e e = \frac{# l i n k s > 0.2 % o f G D P}{# c r e d i t o r \sin d a t a s e t}$
Country-level in-degree can be easily aggregated at the regional level. We refer to the resulting statistic as the interconnectedness of region j:
$I n - \deg r e e_{j} = 100 * \frac{# l i n k s > 0.2 % o f d e b t o r' s G D P i n r g i o n}{# c r e d i t o r s i n d a t a s e t * # d e b t o r s i n r e g i o n_{j}}$
A variant of in-degree is out-degree, which captures the number of recipients a source country is exposed to. The country-level and region-level out-degree are similarly constructed as above, except that the cross-border exposures should be above 0.2 percent of the source ’s GDP to be counted as links.
Concentration in lending relationships is measured by the well-known Herfindahl-Hirschman Index (HHI). Let N be the number of creditors in the dataset and let si be the share of creditor i in country j’s foreign liabilities. The HHI is defined as:
$h_{j} = Σ_{i = 1}^{N} s_{i}^{2}$
and is then normalized to range from 0 to 1:
$H H I = \frac{H - 1 / N}{1 - 1 / N}$
Higher values of the HHI indicate higher concentration of sources. The HHI would approach zero if a recipient has uniform exposures across its N sources. Conversely, it would take the value of one if there is only one active source. We aggregate the HHI at the regional level by taking the average value of the index in the region.

This paper studies cross-border claims of 57 emerging markets and 24 advanced economies (listed in Annex I). Networks were constructed for the period 1999–2009 using the two leading databases used throughout this paper. The analysis on the BIS database was restricted to the 24 source economies that reported data consistently throughout the period 1999–2009. Data from the CPIS on cross-border portfolio holdings is available annually for the period 2001–2009. More than 70 countries reported their external assets to this database but the analysis was restricted to those sources that also reported to the BIS statistics consistently.

There are a few caveats when comparing measures based on CPIS and BIS data. First, the two datasets follow different principles to determine the “location” of the source. In general, the CPIS overemphasizes the importance of financial centers compared to the BIS consolidated statistics. This is because sources operating in a financial center are considered residents in the CPIS, whereas they are consolidated back to their country of origin in the BIS consolidated statistics.³⁴ Second, to the extent that commercial banks hold portfolio investment securities, some overlap may occur. Finally, neither dataset is adjusted for valuation effects.

Interconnectedness and concentration over time. Figure II.2 and II.3 show the in-degree and HH indices over time for EMs and AEs. A few stylized facts stand out:

Interconnectedness has generally increased in the last decade across EMs and AEs as well as asset classes. This trend was partially reversed with the deleveraging during the 2008 crisis—although it may have already resumed in 2009.
Interconnectedness is much stronger for AEs than EMs (note the difference in scale used in Figure II.2). Moreover, the smaller AEs tend to be more interconnected.
In EMs interconnectedness is higher for cross-border bank claims than portfolio claims. There is no clear pattern across asset classes for AEs, though.
Concentration has generally been on a declining trend in the past decade across EMs and AEs as well as asset classes. The most notable exception was cross-border bank claims in European EMs, for which concentration increased rapidly in the years before the crisis.
Concentration across asset classes is still significantly higher in EMs than AEs (again, note the difference in scale in Figure II.3).

Inward vs. outward interconnectedness. Figure II.4 shows the other side of the coin, that is, the interconnectedness index applied to the asset, rather than liability, side of a country; this index is called “out-degree” because it captures a country’s outward interconnectedness—the extent to which a source country’s holdings of cross-border assets are spread out across all the potential recipients. The most striking point is that, whereas AEs’ outward linkages are almost as spread out as inward linkages, for EMs outward linkages are still fairly limited (note again the difference in scale). In other words, interconnectedness in EM economies is generally “one way”, namely it affects mostly their cross-border liability side.³⁵ This observation confirms that, from the perspective of EM economies, the pattern and dynamics of cross-border liabilities is likely to be a far more significant source of cross-border linkages.

Annex III. Top Global Sources and Recipients

Table III.1.

Top Global Sources ^1/

Source: BIS; CPIS; and IMF staff calculations.

^{^1/}Top sources were identified as follows. First, for each recipient country, cross-border sources were ranked according to the size of their bilateral claims to construct a list of the country’s top ten sources. Then, all top sources were ranked based on the frequency they recurred across the lists of country top-sources. This exercise was repeated for each of the three asset classes considered (bank claims, portfolio debt and equity).

Table III.1.

Top Global Sources ^1/

BIS			CPIS: Equity			CPIS: Debt
Creditor	Freq.	Percent	Creditor	Freq.	Percent	Creditor	Freq.	Percent
Germany	80	9.95	United States	69	9.69	Luxembourg	77	9.99
France	77	9.58	Luxembourg	66	9.27	Germany	68	8.82
Netherlands	77	9.58	United Kingdom	61	8.57	United Kingdom	68	8.82
Switzerland	73	9.08	Netherlands	53	7.44	United States	66	8.56
United States	73	9.08	France	47	6.6	France	64	8.3
United Kingdom	71	8.83	Norway	47	6.6	Japan	62	8.04
Japan	60	7.46	Canada	36	5.06	Netherlands	56	7.26
Italy	58	7.21	Ireland	35	4.92	Ireland	38	4.93
Spain	38	4.73	Japan	34	4.78	Italy	38	4.93
Austria	36	4.48	Germany	29	4.07	Denmark	29	3.76
Sweden	32	3.98	Sweden	27	3.79	Austria	28	3.63
Belgium	20	2.49	Switzerland	22	3.09	Switzerland	25	3.24
Ireland	18	2.24	Denmark	16	2.25	Hong Kong	20	2.59
Greece	17	2.11	Korea	14	1.97	Belgium	14	1.82
Denmark	15	1.87	Australia	13	1.83	Cyprus	13	1.69
Canada	14	1.74	Singapore	13	1.83	Singapore	13	1.69
Australia	13	1.62	Brazil	10	1.4	Norway	11	1.43
Portugal	10	1.24	Estonia	10	1.4	Spain	8	1.04
Brazil	8	1	Austria	9	1.26	Greece	8	1.04
Panama	8	1	Chile	9	1.26	Sweden	8	1.04
Turkey	5	0.62	Hong Kong	9	1.26	Canada	7	0.91
Chile	1	0.12	Italy	8	1.12	Panama	7	0.91
			Greece	7	0.98	Russia	7	0.91
			Finland	6	0.84	Finland	6	0.78
			Belgium	5	0.7	Australia	3	0.39
			Spain	5	0.7	Estonia	3	0.39
			Hungary	5	0.7	Korea	3	0.39
			Lebanon	5	0.7	Thailand	3	0.39
			Poland	5	0.7	Costa Rica	2	0.26
			Cyprus	4	0.56	Venezuela	2	0.26
			Latvia	4	0.56	Bulgaria	1	0.13
			Malaysia	4	0.56	Brazil	1	0.13
			Bulgaria	3	0.42	Egypt	1	0.13
			Egypt	3	0.42	Indonesia	1	0.13
			Portugal	3	0.42	Lebanon	1	0.13
			Panama	2	0.28	Latvia	1	0.13
			Russia	2	0.28	Malta	1	0.13
			Turkey	2	0.28	Malaysia	1	0.13
			Uruguay	2	0.28	New Zealand	1	0.13
			South Africa	2	0.28	Pakistan	1	0.13
			India	1	0.14	Portugal	1	0.13
			Kazakhstan	1	0.14	Turkey	1	0.13
			New Zealand	1	0.14	Uruguay	1	0.13
			Romania	1	0.14	South Africa	1	0.13
			Thailand	1	0.14
			Venezuela	1	0.14
Total	804	100.0		712	100.0		771	100.0

Source: BIS; CPIS; and IMF staff calculations.

Table III.1.

Top Global Sources ^1/

BIS			CPIS: Equity			CPIS: Debt
Creditor	Freq.	Percent	Creditor	Freq.	Percent	Creditor	Freq.	Percent
Germany	80	9.95	United States	69	9.69	Luxembourg	77	9.99
France	77	9.58	Luxembourg	66	9.27	Germany	68	8.82
Netherlands	77	9.58	United Kingdom	61	8.57	United Kingdom	68	8.82
Switzerland	73	9.08	Netherlands	53	7.44	United States	66	8.56
United States	73	9.08	France	47	6.6	France	64	8.3
United Kingdom	71	8.83	Norway	47	6.6	Japan	62	8.04
Japan	60	7.46	Canada	36	5.06	Netherlands	56	7.26
Italy	58	7.21	Ireland	35	4.92	Ireland	38	4.93
Spain	38	4.73	Japan	34	4.78	Italy	38	4.93
Austria	36	4.48	Germany	29	4.07	Denmark	29	3.76
Sweden	32	3.98	Sweden	27	3.79	Austria	28	3.63
Belgium	20	2.49	Switzerland	22	3.09	Switzerland	25	3.24
Ireland	18	2.24	Denmark	16	2.25	Hong Kong	20	2.59
Greece	17	2.11	Korea	14	1.97	Belgium	14	1.82
Denmark	15	1.87	Australia	13	1.83	Cyprus	13	1.69
Canada	14	1.74	Singapore	13	1.83	Singapore	13	1.69
Australia	13	1.62	Brazil	10	1.4	Norway	11	1.43
Portugal	10	1.24	Estonia	10	1.4	Spain	8	1.04
Brazil	8	1	Austria	9	1.26	Greece	8	1.04
Panama	8	1	Chile	9	1.26	Sweden	8	1.04
Turkey	5	0.62	Hong Kong	9	1.26	Canada	7	0.91
Chile	1	0.12	Italy	8	1.12	Panama	7	0.91
			Greece	7	0.98	Russia	7	0.91
			Finland	6	0.84	Finland	6	0.78
			Belgium	5	0.7	Australia	3	0.39
			Spain	5	0.7	Estonia	3	0.39
			Hungary	5	0.7	Korea	3	0.39
			Lebanon	5	0.7	Thailand	3	0.39
			Poland	5	0.7	Costa Rica	2	0.26
			Cyprus	4	0.56	Venezuela	2	0.26
			Latvia	4	0.56	Bulgaria	1	0.13
			Malaysia	4	0.56	Brazil	1	0.13
			Bulgaria	3	0.42	Egypt	1	0.13
			Egypt	3	0.42	Indonesia	1	0.13
			Portugal	3	0.42	Lebanon	1	0.13
			Panama	2	0.28	Latvia	1	0.13
			Russia	2	0.28	Malta	1	0.13
			Turkey	2	0.28	Malaysia	1	0.13
			Uruguay	2	0.28	New Zealand	1	0.13
			South Africa	2	0.28	Pakistan	1	0.13
			India	1	0.14	Portugal	1	0.13
			Kazakhstan	1	0.14	Turkey	1	0.13
			New Zealand	1	0.14	Uruguay	1	0.13
			Romania	1	0.14	South Africa	1	0.13
			Thailand	1	0.14
			Venezuela	1	0.14
Total	804	100.0		712	100.0		771	100.0

Source: BIS; CPIS; and IMF staff calculations.

Table III.2.

Top Global Recipients ^1/

Source: BIS; CPIS; and IMF staff calculations.

^{^1/}Top recipients were identified as follows. First, for each source country, cross-border recipients were ranked according to the size of their bilateral claims to construct a list of the country’s top ten recipients. Then, all top recipients were ranked based on the frequency they recurred across the lists of country top-recipients. This exercise was repeated for each of the three asset classes considered (bank claims, portfolio debt and equity).

Table III.2.

Top Global Recipients ^1/

BIS			CPIS: Equity			CPIS: Debt
Debtor	Freq.	Percent	Debtor	Freq.	Percent	Debtor	Freq.	Percent
UK	22	9.58	United States	54	9.93	United States	52	9.56
United States	22	9.58	UK	51	9.38	UK	50	9.19
Germany	20	8.71	Luxembourg	45	8.27	Germany	47	8.64
France	19	8.27	France	42	7.72	France	46	8.46
Netherlands	18	7.84	Germany	36	6.62	Netherlands	45	8.28
Ireland	12	5.22	Ireland	36	6.62	Ireland	32	5.88
Italy	12	5.22	Switzerland	27	4.96	Luxembourg	28	5.15
Spain	11	4.78	Japan	24	4.41	Spain	23	4.23
Luxembourg	11	4.78	Netherlands	24	4.41	Italy	21	3.86
Canada	8	3.48	Australia	17	3.13	Australia	16	2.95
Japan	6	2.61	Spain	17	3.13	Brazil	16	2.95
Belgium	5	2.17	Canada	15	2.76	Japan	14	2.58
Brazil	4	1.74	Brazil	12	2.21	Austria	13	2.40
Korea	4	1.74	Italy	12	2.21	Greece	12	2.20
Poland	4	1.74	China	11	2.02	Belgium	9	1.65
Austria	3	1.30	Hong Kong	10	1.84	Sweden	9	1.65
Switzerland	3	1.30	Russia	9	1.65	Canada	8	1.47
Chile	3	1.30	Belgium	7	1.29	Korea	7	1.28
China	3	1.30	Korea	7	1.29	Switzerland	6	1.10
Hungary	3	1.30	Singapore	7	1.29	Denmark	6	1.10
Mexico	3	1.30	Sweden	7	1.29	Norway	6	1.10
Argentina	2	0.87	Austria	4	0.74	Hong Kong	5	0.92
Australia	2	0.87	Greece	4	0.74	India	5	0.92
Denmark	2	0.87	Mexico	4	0.74	Finland	4	0.73
Finland	2	0.87	Argentina	3	0.55	Hungary	4	0.73
Hong Kong	2	0.87	Cyprus	3	0.55	Mexico	4	0.73
Norway	2	0.87	Hungary	3	0.55	Poland	4	0.73
Romania	2	0.87	India	3	0.55	Portugal	4	0.73
Sweden	2	0.87	Malaysia	3	0.55	Argentina	3	0.55
Bulgaria	1	0.43	Panama	3	0.55	Chile	3	0.55
Costa Rica	1	0.43	Poland	3	0.55	Malaysia	3	0.55
Dominican Rep.	1	0.43	Portugal	3	0.55	Russia	3	0.55
Estonia	1	0.43	Turkey	3	0.55	Singapore	3	0.55
Greece	1	0.43	Bulgaria	2	0.37	China	2	0.37
Guatemala	1	0.43	Chile	2	0.37	Colombia	2	0.37
Croatia	1	0.43	Denmark	2	0.37	Cyprus	2	0.37
India	1	0.43	Egypt	2	0.37	Sri Lanka	2	0.37
Lithuania	1	0.43	Estonia	2	0.37	Lithuania	2	0.37
Latvia	1	0.43	Finland	2	0.37	Peru	2	0.37
New Zealand	1	0.43	Indonesia	2	0.37	El Salvador	2	0.37
Panama	1	0.43	Jordan	2	0.37	Thailand	2	0.37
Peru	1	0.43	Norway	2	0.37	Venezuela	2	0.37
Portugal	1	0.43	Peru	2	0.37	South Africa	2	0.37
Russia	1	0.43	Romania	2	0.37	Bolivia	1	0.18
Singapore	1	0.43	Thailand	2	0.37	Egypt	1	0.18
Serbia	1	0.43	Uruguay	2	0.37	Estonia	1	0.18
Turkey	1	0.43	Ecuador	1	0.18	Guatemala	1	0.18
			Israel	1	0.18	Indonesia	1	0.18
			Lebanon	1	0.18	Iceland	1	0.18
			New Zealand	1	0.18	Jamaica	1	0.18
			Pakistan	1	0.18	Macedonia	1	0.18
			Serbia	1	0.18	New Zealand	1	0.18
			Ukraine	1	0.18	Panama	1	0.18
			Venezuela	1	0.18	Philippines	1	0.18
			Vietnam	1	0.18	Romania	1	0.18
						Tunisia	1	0.18
						Turkey	1	0.18
Total	230	100.0		544	100.0		545	100.0

Source: BIS; CPIS; and IMF staff calculations.

Table III.2.

Top Global Recipients ^1/

BIS			CPIS: Equity			CPIS: Debt
Debtor	Freq.	Percent	Debtor	Freq.	Percent	Debtor	Freq.	Percent
UK	22	9.58	United States	54	9.93	United States	52	9.56
United States	22	9.58	UK	51	9.38	UK	50	9.19
Germany	20	8.71	Luxembourg	45	8.27	Germany	47	8.64
France	19	8.27	France	42	7.72	France	46	8.46
Netherlands	18	7.84	Germany	36	6.62	Netherlands	45	8.28
Ireland	12	5.22	Ireland	36	6.62	Ireland	32	5.88
Italy	12	5.22	Switzerland	27	4.96	Luxembourg	28	5.15
Spain	11	4.78	Japan	24	4.41	Spain	23	4.23
Luxembourg	11	4.78	Netherlands	24	4.41	Italy	21	3.86
Canada	8	3.48	Australia	17	3.13	Australia	16	2.95
Japan	6	2.61	Spain	17	3.13	Brazil	16	2.95
Belgium	5	2.17	Canada	15	2.76	Japan	14	2.58
Brazil	4	1.74	Brazil	12	2.21	Austria	13	2.40
Korea	4	1.74	Italy	12	2.21	Greece	12	2.20
Poland	4	1.74	China	11	2.02	Belgium	9	1.65
Austria	3	1.30	Hong Kong	10	1.84	Sweden	9	1.65
Switzerland	3	1.30	Russia	9	1.65	Canada	8	1.47
Chile	3	1.30	Belgium	7	1.29	Korea	7	1.28
China	3	1.30	Korea	7	1.29	Switzerland	6	1.10
Hungary	3	1.30	Singapore	7	1.29	Denmark	6	1.10
Mexico	3	1.30	Sweden	7	1.29	Norway	6	1.10
Argentina	2	0.87	Austria	4	0.74	Hong Kong	5	0.92
Australia	2	0.87	Greece	4	0.74	India	5	0.92
Denmark	2	0.87	Mexico	4	0.74	Finland	4	0.73
Finland	2	0.87	Argentina	3	0.55	Hungary	4	0.73
Hong Kong	2	0.87	Cyprus	3	0.55	Mexico	4	0.73
Norway	2	0.87	Hungary	3	0.55	Poland	4	0.73
Romania	2	0.87	India	3	0.55	Portugal	4	0.73
Sweden	2	0.87	Malaysia	3	0.55	Argentina	3	0.55
Bulgaria	1	0.43	Panama	3	0.55	Chile	3	0.55
Costa Rica	1	0.43	Poland	3	0.55	Malaysia	3	0.55
Dominican Rep.	1	0.43	Portugal	3	0.55	Russia	3	0.55
Estonia	1	0.43	Turkey	3	0.55	Singapore	3	0.55
Greece	1	0.43	Bulgaria	2	0.37	China	2	0.37
Guatemala	1	0.43	Chile	2	0.37	Colombia	2	0.37
Croatia	1	0.43	Denmark	2	0.37	Cyprus	2	0.37
India	1	0.43	Egypt	2	0.37	Sri Lanka	2	0.37
Lithuania	1	0.43	Estonia	2	0.37	Lithuania	2	0.37
Latvia	1	0.43	Finland	2	0.37	Peru	2	0.37
New Zealand	1	0.43	Indonesia	2	0.37	El Salvador	2	0.37
Panama	1	0.43	Jordan	2	0.37	Thailand	2	0.37
Peru	1	0.43	Norway	2	0.37	Venezuela	2	0.37
Portugal	1	0.43	Peru	2	0.37	South Africa	2	0.37
Russia	1	0.43	Romania	2	0.37	Bolivia	1	0.18
Singapore	1	0.43	Thailand	2	0.37	Egypt	1	0.18
Serbia	1	0.43	Uruguay	2	0.37	Estonia	1	0.18
Turkey	1	0.43	Ecuador	1	0.18	Guatemala	1	0.18
			Israel	1	0.18	Indonesia	1	0.18
			Lebanon	1	0.18	Iceland	1	0.18
			New Zealand	1	0.18	Jamaica	1	0.18
			Pakistan	1	0.18	Macedonia	1	0.18
			Serbia	1	0.18	New Zealand	1	0.18
			Ukraine	1	0.18	Panama	1	0.18
			Venezuela	1	0.18	Philippines	1	0.18
			Vietnam	1	0.18	Romania	1	0.18
						Tunisia	1	0.18
						Turkey	1	0.18
Total	230	100.0		544	100.0		545	100.0

Source: BIS; CPIS; and IMF staff calculations.

Annex IV. Importance of Global Factors in Driving Capital Inflows to EMs³⁶

This annex investigates the quantitative importance of global factors and country-specific fundamentals in driving cross-border capital inflows to EMs. The conservative estimates discussed below suggest that global factors could explain at least 25 percent of variations in total gross inflows; for portfolio inflows, however, global factors are likely much more important (above 50 percent). Moreover, the relative importance of global factors is likely to shift over time, reflecting in particular global financial and liquidity conditions.

These shares were calculated using the same panel framework as in IMF, 2011e, which included both global and domestic variables as determinants of capital inflows. One problem in computing the contribution of global factors in this framework is that the overall variance of inflows (in log levels) in a panel of 48 EMs is largely driven by cross-country variation of inflows. The latter is often a result of country specific factors (e.g. political conditions or idiosyncratic developments such as privatization or lumpy investment projects) that are likely to drive capital flows without being directly linked to differences in domestic

macroeconomic conditions—that is, the domestic pull factors that one would like to isolate here.³⁷ To correct for this potential bias, the total variance of underlying flows was broken down into two components, variance across countries and over time. The contribution of global factors was then computed as the ratio of their variance to the variance of gross capital inflows over time.

In this framework, global factors (U.S. long term interest rates and the VIX risk index) explained on average around 25 percent of the cross-time variation of gross total capital inflows to EMs; for portfolio inflows, the estimated share rose to 54 percent. These shares were computed by treating domestic variables as completely unaffected by global variables, while in practice they are likely to be influenced by the latter. When domestic variables were omitted from the estimated regressions, the share of cross-time variation explained by global factor rose to 65 and 87 percent, respectively, for total and portfolio inflows—an indirect sign that domestic variables are indeed affected significantly by global factors. These estimates can be thought of as an upper bound for the role of global factors for capital inflows.

Robustness checks were performed by estimating the same regressions discussed in the previous paragraph for seven countries separately (Table IV.1). The results confirmed that global factors explain on average a fifth of the variation of total inflows for the seven countries. For portfolio inflows, the average share increased to 48 percent. Thus, these estimates were broadly aligned with the average shares derived from the panel approach.

Table IV.1:

Share of Global Factors in Explaining Gross Capital Inflows

As a further robustness check, the methodology used in IMF, 2011b, was applied to the current sample of 48 EMs for both gross and net flows (normalized as a percent of GDP). More specifically, this entailed three steps:

Step 1. For each period t, capital flows were decomposed into two parts, Y(i,t)=a(t) + error(i, t), where error(i, t) denotes the part attributable to country-specific factors and a(t) the part driven by global factors that are constant across countries but vary over time. The term a(t) is estimated for each period t as a cross-section average of flows. The Residual Sum of Squares (RSS) was then calculated for each period.
Step 2. The Total Sum of Squares (TSS) was derived for each period as the crosscountry sum of squared differences of flows from their sample mean.
Step 3. The R-squared was calculated as 1-(RSS/TSS) for each period. This yielded the share of variance attributable to global factors.

The results are shown in Figure IV.1 and IV.2 for net and gross inflows, respectively. Global factors explained an important, though relatively small, share of total capital inflows—less than 20 percent for net inflows and less than a third for total gross inflows. This exercise, however, underscored how relative importance of global factors shifts over time and across different types of flows.

Annex V. Construction of a Total Spillover Index³⁸

This annex discusses the derivation of “total spillover index” following the methodology introduced by Diebold and Yilmaz (2009, 2010, and 2011).

This index is derived in two steps from the forecast variance decomposition at a given horizon in an N-variables vector autoregression (VAR) model. First, for each variable i in the model, the share of its forecast error variance decomposition at a given horizon coming from shocks to variable j is added for all i≠j. This yields the “Total directional connectedness from other variables” for variable i. Then, the “Total directional connectedness from other variables” are added across all variables, i = 1,…,N , to generate the total spillover index. Intuitively, this index captures the average share of forecast variance explained by shocks in other parts of the system. To build the time series of the total spillover index, the exercise was repeated by rolling the VAR estimation over sample windows.

For the application on financial variables shown in the paper, an unrestricted VAR was estimated over the period 1990:01 to 2010:12. The estimated model included five variables³⁹: the VIX index, the first difference of the log real S&P 500 index, the first difference of the log real commodity price index, the U.S. federal funds rate, and the aggregate real return of EMs stock markets, which is constructed from the z-scores associated with the first principal component of a sample of EMs real stock returns.⁴⁰ The VAR’s number of lags (two) was determined using the sequential modified likelihood ratio test, the Akaike-Schwarz based lower maximum likelihood criteria. The Portmanteau autocorrelation test, the normality test, and the White heteroskedasticity test performed on the residuals were used as specification tests. A recursive Cholesky orthogonalization of the error terms for the variance decomposition analysis was then used to identify structural shocks⁴¹. Finally, error forecast variance at a 6 months horizon-step ahead was used to construct a full-sample “Connectedness table” (Table V.1).

The lines of the table (except the last line and excluding the last column “total connectedness from others”) provide the variance decomposition for each variable at a 6 months horizon; by construction, this sum equals to 100 percent. The last column of the table sums off-diagonal elements and provides information about the relative importance of other random innovations in affecting a specific variable in the model. For instance, for the S&P 500, the value of 64 indicates that 64 percent of its 6-month-ahead error forecast variance is due to shocks to the other variables included in the VAR. This total contribution from other innovations is denoted “Total directional connectedness from other variables”. Similarly, the last line of the table gives the “total directional connectedness to other variables”; this is the sum across variables of the contribution of a specific innovation to the variation of other variables and identifies the “connectedness” the variable associated with that innovation provides to the other variables in the model.

Table V.1.

Connectedness

Source: Datastream and IMF staff calculations.

Table V.1.

Connectedness

	VIX	S&P 500	Commodit y Price Index	Fund Rate	EM real stock return	Total directional Connectedness from others
VIX	95.16	1.26	2.66	0.22	0.70	4.84
S&P 500	61.00	36.43	0.87	1.47	0.23	63.57
Commodity Price Index	20.76	1.11	77.20	0.07	0.85	22.80
Fund Rate	25.39	0.25	1.61	71.73	1.02	28.27
EM real stock return	2.27	4.36	1.21	2.04	90.12	9.88
Total directional
Connectedness to others	109.42	6.99	6.35	3.80	2.79	129.35

Source: Datastream and IMF staff calculations.

Table V.1.

Connectedness

	VIX	S&P 500	Commodit y Price Index	Fund Rate	EM real stock return	Total directional Connectedness from others
VIX	95.16	1.26	2.66	0.22	0.70	4.84
S&P 500	61.00	36.43	0.87	1.47	0.23	63.57
Commodity Price Index	20.76	1.11	77.20	0.07	0.85	22.80
Fund Rate	25.39	0.25	1.61	71.73<

In-degree	14.33* (7.66)
HHI		-11.64* (6.62)
Domestic demand growth in AE trading partners	1.12* (0.57)	1.36** (0.58)
Foreign bank claims (percent of GDP, in logs)	-2.44*** (0.81)	-1.44 (0.86)
GIR in percent of (short-term debt at residual maturity plus current account deficit, in logs)	3.11*** (0.90)	2.10** (0.97)
Observations	40	40
R-squared	0.53	0.53

In-degree	14.33* (7.66)
HHI		-11.64* (6.62)
Domestic demand growth in AE trading partners	1.12* (0.57)	1.36** (0.58)
Foreign bank claims (percent of GDP, in logs)	-2.44*** (0.81)	-1.44 (0.86)
GIR in percent of (short-term debt at residual maturity plus current account deficit, in logs)	3.11*** (0.90)	2.10** (0.97)
Observations	40	40
R-squared	0.53	0.53

Topics

Countries

Series

About

Resources

Abstract

I. Executive Summary

II. Context and Motivation1

III. Snapshot of Cross-Border Financial Linkages

IV. Cross-Border Financial Linkages and Shock Transmission

V. What Explains Cross-Border Linkages?

VI. Conclusions

Annex I. Country Groupings

Annex II. Network Analysis27

Insights from network theory

Measuring interconnectedness and concentration

Annex III. Top Global Sources and Recipients

Annex IV. Importance of Global Factors in Driving Capital Inflows to EMs36

Annex V. Construction of a Total Spillover Index38

II. Context and Motivation¹

Annex II. Network Analysis²⁷

Annex IV. Importance of Global Factors in Driving Capital Inflows to EMs³⁶

Annex V. Construction of a Total Spillover Index³⁸