Book
Share
Chapter

Chapter 13 Real and Financial Vulnerabilities from Cross-Border Banking Linkages

Author(s):
Li Lian Ong, and Andreas A. Jobst
Published Date:
September 2020
Share
  • ShareShare
Show Summary Details
Author(s)
Kyunghun Cerutti and Srobona Mitra

This chapter is based on IMF Working Paper 14/136 (Kim and Mitra 2014). The authors would like to thank, without implicating, James Morsink, who suggested the project and provided extensive and useful comments; Eugenio Cerutti, Jiaqian Chen, Allison Holland, Gian Maria Milesi-Ferretti, Prachi Mishra, Lam Nguyen, Ratna Sahay, Jason Weiss, and Hongyan Zhao for other helpful suggestions; and Juan Sole and Marco Espinosa-Vega for making available the codes for network analysis through an Excel add-in; and various participants in internal seminars at the IMF.

This chapter looks at the vulnerabilities stemming from banking sector linkages between countries and their macroeconomic effects. It finds that credit risks (from a banking system’s claims on other countries) and funding risks (from a banking system’s liabilities to another) declined between 2008 and 2012. It also finds that funding vulnerabilities have real effects. During normal times, funding vulnerabilities are associated with significant positive GDP growth surprises. During crisis times, funding vulnerabilities are associated with significant negative GDP growth surprises. The results imply that policymakers should pay more attention to understanding cross-border funding risks.

1. Introduction

The global financial crisis made it clear that financial shocks could be quickly transmitted through global banks. The tightly interconnected financial systems were put through several tests during the crisis. The banking linkages, by far the largest and the deepest segment of financial flows, saw reduced flows.

Against this backdrop, this chapter asks two questions. First, how have countries’ vulnerabilities arising from banking network linkages changed? The chapter examines two kinds of risks— credit risk and funding risk. These risks are related to the nature of the interlinkages— credit risks materialize through a banking system’s claims on other countries, and funding risks arise through banking system’s liabilities to one another. The vulnerabilities are related to both exposures to these risks and the capital buffers available against these risks. Second, what are the macroeconomic effects of these vulnerabilities? That is, are these specific vulnerabilities associated with real GDP growth beyond what is expected in macroeconomic forecasts?

The chapter explores the financial risks of cross-border banking linkages using network analysis. Rather than just identifying and quantifying linkages, the chapter simulates the impact on capital levels of the credit and funding shocks that could be transmitted through direct and indirect (domino effect) banking linkages. The chapter examines whether the potential impact on capital— summarized by vulnerability indices— has changed in the last five years. Using network analysis (Espinosa-Vega and Solé 2011), the chapter shows the trends in the financial systems’ vulnerability to network effects of shocks on either side of the balance sheet.

The chapter then asks whether the vulnerability of a banking system from interconnections influences output. For the network analysis to have macro-financial implications, the real effects of higher vulnerability to network shocks are estimated using an econometric model. Specifically, a set of panel fixed effect regressions examine the relationship between vulnerability to cross-border credit or funding shocks and GDP growth rate surprises, measured by the difference between actual GDP growth and Consensus Forecasts.

The chapter has two main findings. First, vulnerabilities of banking systems to both credit and funding risks have declined since the crisis. This decline is due to both lower exposures and increases in capital for the global banking system. Second, funding vulnerabilities have real effects. During normal times, funding vulnerabilities are positively associated with GDP growth surprises; during crises, the same vulnerabilities exacerbate the negative GDP growth surprises. Credit vulnerabilities, on the other hand, are not associated with GDP surprises.

The rest of the chapter is organized as follows. The related literature is discussed in Section 2; the methodology and the data in Section 3; the findings on the vulnerability trends and the association of the vulnerabilities with GDP growth surprises are discussed in Sections 4 and 5, respectively; and Section 6 concludes.

2. Related Literature

The chapter builds on the recent literature on cross-border financial interconnectedness and its implications for financial stability and real output. Kalemli-Ozcan, Papaioannou, and Perri 2013 find that higher banking linkages are associated with more divergent output cycles during normal times; however, this relationship becomes weaker during financial crisis. Abiad and others 2013 distinguish between traditional financial linkages and common shocks to show that output comovement across countries— synchronized output collapses— occurs during financial crises through common shocks. Cetorelli and Goldberg 2010 show how the US financial crisis was transmitted to other countries through the relationship between multinational banks and their foreign affiliates. Albertazzi and Bottero 2014 suggest that the foreign banks restricted credit supply more than their domestic counterparts, using disaggregated Italian bank-firm data. De Haas and Van Lelyveld (2014), Giannetti and Laeven (2012), and Popov and Udell (2012) have empirical evidence to show that multinational banks restricted credit supply in the host countries during the financial crisis.

Cihák, Munoz, and Scuzzarella 2011 show an M- shaped relationship between the financial stability of a country’s banking sector and its cross-border interconnectedness measured by network centrality measures— starting from low integration, increases in global interconnectedness for the banking system are associated with a reduced probability of a banking crisis. For a banking system whose interconnect-edness is over a certain value, increases in interconnectedness can increase the probability of a banking crisis. Relatedly, Minoiu and others 2015 show that increases in a country’s own connectedness and decreases in its neighbors’ connectedness are associated with a higher probability of banking crises. Nier and others 2007 investigate how systemic risk is affected by the structure of the banking system, using network models.

Espinosa-Vega and Solé 2011 show that network analysis can be used as a tool for cross-border financial surveillance. By simulating credit or funding shocks, they obtain vulnerability indices for each banking system. Using this tool, Espinosa-Vega, Solé, and Kahn (IMF 2010) also propose a framework for capital requirements for those banks that have a large contribution to systemic risk in a network. Cerutti, Claessens, and McGuire (2011) highlight data needed for properly analyzing contagion risk, an exercise similar in spirit to the network analysis, and Cerutti 2013 proposes two new measures for better capturing creditor banking systems’ foreign credit exposures and borrower countries’ reliance on foreign bank credit, by combining Bank for International Settlements (BIS) data with bank-level data.

This is the first work to distinguish between cross-border risks arising from the asset and the liability side of the banking system’s balance sheet and relate these different risks to macroeconomic effects. The chapter applies the methodology proposed by Espinosa-Vega and Solé (2011) to a dataset that covers 20 countries over 2006–12 and shows the real impacts for the countries receiving the shocks. The chapter documents how the cross-border vulnerabilities of the banking system have evolved since 2006 and shows how the vulnerability index from the network analysis is associated with output shocks during normal times and crises.

3. Data and Methodology

The vulnerability from interconnections goes beyond the simple mapping of exposures between countries. The vulnerability or susceptibility to network effects is measured by the potential capital shortfall in the event of a tail risk in which one banking system fails. It is measured by the average change in the capital level in percentage of the pre-shock capital due to the direct and domino effects of every banking system failing. Therefore, the vulnerability of any country to a shock in another banking system depends upon four factors: effects through direct bilateral links, domino effects through indirect network links, own capital levels, and capital levels in the major shock-propagating countries. Vulnerability goes up with stronger banking bilateral links and gets magnifed by domino effects running through link-of-links. Lower capital buffers in the shock recipients, as well as in shock propagators, increases vulnerability in any given country. Of course, the use of aggregate data might not capture potential systemic vulnerabilities arising from individual large institutions.

Data

To run the network analysis, data is needed on the matrix of exposures between countries. This means a banking system’s credit (claims) and liabilities vis-à- vis another country’s banking system are needed. The BIS consolidated banking statistics is used for the purpose. Since it does not have data on cross-border liabilities of banking systems, this is proxied by looking at the claims of the counterparty banking systems. The liabilities side, therefore, is measuring the liabilities of all sectors of the economy to BIS reporting banks with headquarters in another country. Even though it is imprecise, it is assumed that most of these liabilities are sourced through the banks and measure the banking system’s indirect liabilities to the BIS reporting banks in the other country. This is the best that can be done with the published data, which is available for 20 countries.1

There is a 20 by 20 matrix for each of the years 2005 through 2012. For instance, in 2008, the US banks lent $268 billion to the United Kingdom and the United States (all sectors) borrowed $1.217 trillion from the United Kingdom. By 2012, the United States lent more than twice to the United Kingdom and borrowed less from the Kingdom (Table 13.1).

Table 13.1.Capital and Financial Exposure between Banking Systems (In millions of US dollars, column countries’claims on rows)
2008:Q4
CapitalCountryAUATBECAFlFRDEGRINIEITJPNLPTESSECHTRGBUS
41,858Australia2124506510350NaN43577509613552277271409520827256259335972572215972410032445287
56,095Austria119151171358NaN205221056791732315806146042675792038403100181513724425124874550
39,476Belgium277929855055NaN11113541295173997820311687210911523641828122034122180677784350026415
59,702Canada643918497263NaN2615541701141230712385249645193397312841656253019071198075666152
3,615Finland50310682746218667614044726NaN1665635740214717551073802945264093880
368,163France107381087912564412430NaN19324667386325496552531245621246157261463261139966233126024110777068
183,041Germany11372524275889312242NaN279538222819724971433744715833417413898544711874498113691519015997892620
15,083Greece480561710175395NaN7522438389198480951361761286863761012131069552135127136753
52,769India14737615397NaNNaN113551949840NaN76413168220201571155307531904967238313
30,583Ireland165150974555014233NaN681152022023232632443923857354383781148325247203879819044033014
154,393Italy1049017628519512334NaN4688502071942784864653748270669553483486803912199467637483925526
363,573Japan239748627976225NaN218920656191271617228NaN28170211214943121091211113158123333
108,956Netherlands6998120688545310545NaN128186167279807964175502436445822898521375935251784284012960152599
18,026Portugal318253912040NaNNaN29918444924049634161973056138427742456975243219521848
141,955Spain21077919439643050NaN1764212536762651663370428463257111247732865571062036024512457233458
54,611Sweden725161125811349NaN161543793517170617319881450581979011917781080164258323
105,385Switzerland35511092398403877NaN574836746945430774151134224846188792250543655454774490922865
18,582Turkey104251715610NaNNaN123551607218317105NaNNaN33862122990613130041591734012806
468,068United Kingdom1039582356012777464156NaN39455750913369653600222201498981640721806077639349916387132192023418268187
1,088,470United States4193121060113161430465NaN766345640501395377891224773271191164233592088821326234270382713351361217127
2012:Q4
CapitalCountryAUATBECAFlFRDEGRINIEITJPNLPTESSECHTRGBUS
43,058Australia79219572221936918053237246910437152265130606793322425152352282062872924115419
71,233Austria279730109632014618756859635731510114561709839124482419338570198781911998
48,663Belgium882155023315812229832792427611535244427192911166184415226237863741631801618469
103,774Canada1883898213401501690127668166279937138336263910056168169524672186011104838129360
5,107Finland864869452194874341840416834NaN120044005687552014151832451511026312242
462,704France82511172725319272362715195139167075452274576416636867377699031531803057918957221012213807
237,477Germany21718437461274925595270619764335102078229923707314581118500721595739580606717853057273571217456
8,310Greece9533132NaNNaN279852934113902404234374007797615279056313201
93,048India96063824045852NaN15422235811NaN22972542213876132902561093868426480077
31,989Ireland25311399200004591399379548158140510110207234861272639786047156014735212197546515
174,556Italy578156751033946362743432071292005201469553706833068283927740143319981624922742716
275,686Japan26875NaN7951574529268945681104578184NaN802922309295668161913130355372517
153,016Netherlands9960812723600132822295158134157528309513812235198106525093921935110501352142636171369106970
22,274Portugal132791716NaN139169162167017404731725117845007156719015030173374765
189,093Spain8913018947125134331080331207172181034085229772122153686226742792175502088286349436
61,728Sweden2525153960527273462213033487388157530216720939755715925949583131547027889
159,268Switzerland822096961323562770168058606458546558491079029162136452065734731852718646576823
36,178Turkey380181113212517NaN308471901831083107NaN631383902078222145521561473752824492
761,648United Kingdom13342717012258751076942312224666409259118665382111414497701889091289065246406941488721663893090634309
1,622,337United States1078141076020742720340461402553496792423693897107309771296100164675503820357610315067062948191080697
Sources: Bank for International Settlements; Bankscope; and authors’calculations.Note: AT = Austria; AU = Australia; BE = Belgium;CA = Canada; CH = Switzerland; DE = Germany; ES = Spain; Fl = Finland; FR= France;GB = United Kingdom;GR = Greece; IE = Ireland; IN = India; IT= Italy; JP = Japan; NL= Netherlands; PT= Portugal; SE = Sweden;TR = Turkey; US = United States.
Sources: Bank for International Settlements; Bankscope; and authors’calculations.Note: AT = Austria; AU = Australia; BE = Belgium;CA = Canada; CH = Switzerland; DE = Germany; ES = Spain; Fl = Finland; FR= France;GB = United Kingdom;GR = Greece; IE = Ireland; IN = India; IT= Italy; JP = Japan; NL= Netherlands; PT= Portugal; SE = Sweden;TR = Turkey; US = United States.

In order to understand the vulnerabilities from cross-border exposures, one needs to weigh the exposures against the financial buffers. So, data on capital was needed, which was obtained from Bankscope. The sum was taken of the capital that each banking system’s commercial banks, savings banks, cooperative banks, real estate and mortgage banks, investment banks, other nonbanking credit institutions, and specialized governmental credit institutions own. A wide net was cast to capture data on capital from as many institutions residing and headquartered in a country as was possible to get a sense of buffers.

Methodologies

Deriving vulnerability indices based on network analysis

The network model used in this chapter was developed in Chapter 2 of the IMF’s April 2009 Global Financial Stability Report (IMF 2009) and described in Espinosa-Vega and Solé 2011. The model runs simulations using the data on exposures and capital. Specifically, it lets each banking system fail and calculates the impact of the credit risk from such a failure on other banking systems’ capital. Similarly for funding risk. There are both direct and domino effects of a banking system’s failure on others.

The method can be illustrated by means of a stylized balance sheet of a banking system, say A (Figure 13.1). For credit risk (panel 1), if, another banking system B’s banks fail due to some unexplained event, it is unable to repay λ (the assumption is 0.5 in the baseline) of its dues to all other countries. These assets then go “bad” for all the creditor banking systems, A is one of them, and these should have sufficient capital to absorb this loss. If they don’t, then the banking systems are said to fail, and these then trigger domino impacts on all others. The simulation goes on until there are no more failures.

Credit Shock and Funding Shock Illustrated with Stylized Banking System Balance Sheets

Source: April 2009 Global Financial Stability Report, Chapter 2 (IMF 2009); and Sole and Espinosa-Vega 2010.

Note: A “*” represents the amount by which capital, k, will be hit in the first round. x = cross-border credit and funding; a = other assets; d = other liabilities, like customer deposits and debt; k = capital; λ = fraction of interbank loans that does not get repaid (0.50 in the baseline); ρ = fraction of interbank liabilities that does not get rolled over (0.50 in the baseline); δ = haircut on interbank assets that need to be fire sold to replace the fraction of interbank funding that is not rolled over (1 in the baseline).

For funding risk (panel 2), if B fails, it is unable to roll over ρ (the assumption is 0.5 in the baseline) times other countries’ liabilities, including A’s. A, and other countries, then try to fire sell their assets at a haircut (the assumption is half, which translates into (δ= 1) and take a hit on capital. If it fails, this triggers further failures. Again, the domino goes on until there are no more failures.

The network model produces vulnerability indices. The index is simply the average capital depletion if other banking systems fail. This number is derived by running the network model for each country, at each point of time, 2005–12, separately for credit risk and funding risk. So, there is a credit vulnerability index and a funding vulnerability index for each country. Then there is a global index (for all 20 countries) that takes a weighted average of the indices for each country, weighted by the sum of gross credit and liabilities of each country.

The vulnerability index has a practical meaning. The credit index indicates the potential capital loss (in percent of preshock capital) of a banking system’s opening up to foreign expansions, increasing foreign claims, or not having adequate capital buffers against those claims. The funding index gives information on the potential capital loss rate of a banking system due to opening up to higher foreign funding (liabilities risk) without adequate capital buffers to withstand fire sales if necessary. The index itself is influenced by four factors for given levels of the parameters, λ, ρ, and δ: direct linkages, indirect linkages, own capital levels, and those of others.

Panel fixed effects regressions are used to look at the association between GDP growth surprises and the vulnerability indices, for 20 countries, for seven years 2006–12. The GDP growth surprises are calculated by taking the difference between actual GDP growth and the forecast of GDP growth made in the previous December by consensus economics. The average growth surprises for the 20 countries show the large negative surprises during the crisis years 2008 and 2009 (Figure 13.2). The regressions take the growth surprise as the dependent variable, and regresses it on a dummy variable that takes the value of 1 for the two crisis years, the vulnerability index, and a term that interacts the vulnerability index with the crisis dummy (see equation 13.1).

If the cross-border credit and funding risks are well understood by macroeconomic forecasters, the indices would not be expected to affect the growth surprises. This is because the GDP forecasts would already take into account the risks that could affect a country through the cross-border banking channels so that the residuals, the GDP growth surprises, should not be correlated with information available at the time of making these forecasts.

Growth Rate Surprise

(Average difference between actual and forecasted GDP for 20 countries, in percentage points)

Source: Consensus Forecasts.

Note: GDP growth rate surprise = actual GDP growth rate (WEO) – GDP growth rate forecast (Consensus Forecasts, average of the GDP growth rate forecasted over the previous December).

To check if data on overall exposures (foreign claims + foreign liabilities) and capital, separately would have delivered similar results, obviating the need to run the network analysis, a second set of regressions using these components was added, instead of the vulnerability indices (equation 13.2). If higher exposures and lower capital helped explain growth surprises, then understanding these components of the network analysis would be beneficial by themselves.

4. Is the World Safer from Cross-Border Banking Linkages?

The matrix of banking exposures across countries reveals notable changes between 2008 and 2012 (Table 13.1). The financial exposures and funding of non-European countries are on the rise, especially of Canada, Japan, and the United States. The euro area countries have all seen a drop in both cross-border exposures and funding; this is especially so for France and Germany. This phenomenon, often called “fragmentation,” has left policymakers worried about the cost of funds and the availability of credit in European countries. Whether the world is safer from cross-border banking connections depends upon bilateral exposures, network exposures through domino effects, and on own and other countries’ capital levels.

Vulnerability of the overall global banking system to network shocks was high before 2008 (Figures 13.3 and 13.4). Going back to 2006, about 25–30 percent of capital, on an average in a country could have been impaired due to network effects of credit and funding shocks. Since then, countries’ susceptibility to these shocks started coming down until 2008, and then fell after that. The decrease until 2008 was mostly due to the lower volume of flows between advanced countries since mid-2007. The vulnerabilities in 2006, based on published balance sheet data on the banking network, could have served as an early warning on the extent of losses that banking systems would suffer if there were to be an extreme event.

Vulnerability to Credit Shock1

(Financial exposure weighted average of vulnerability; global banking system, 2005–12)

Sources: Bank for International Settlements; Bankscope; and authors’ estimates.

1 The index of vulnerability shows the percentage of capital impairment in a banking system due to the failure of other banking systems. The aggregate index shown above is the weighted average of the vulnerability indices of the 20 countries in the sample, weighted by the country’s total financial exposure.

Since the end of 2008, banking systems have been generally less vulnerable to ripple effects from network shocks due to two reasons. With the collapse of Lehman Brothers and the severance in some linkages due to the materialization of the adverse shocks, individual banking systems now had lower volume of inflows through banks. And capital levels had increased on the aggregate after the crisis so that for any inflow the buffers were greater across countries, in general, to absorb the shocks.

To show that higher buffers were not entirely responsible for the lower vulnerability levels, the network analysis is repeated for 2009–12, assuming that the capital levels are constant at the 2008 levels (Figures 13.3 and 13.4). Even after adjusting for capital, the vulnerability indices (weighted by total exposures of countries) have trended down for both credit and funding shocks, which suggests that the actual strength and number of interconnections had also fallen.

Vulnerability to a Funding Shock1

(Financial exposure weighted average of vulnerability; global banking system, 2005–2012)

Sources: Bank for International Settlements; Bankscope; and authors’ estimates.

1 The index of vulnerability shows the percentage of capital impairment in a banking system due to the failure of other banking systems. The aggregate index shown above is the weighted average of the vulnerability indices of the 20 countries in the sample, weighted by the country’s total financial exposure.

The aggregate results mask wide cross-country differences in vulnerability trends on credit shocks. There are three groups of countries depending upon whether vulnerabilities on cross-border assets have trended down or up or largely remained unchanged between 2008 and 2012 (Figure 13.5):

  • Belgium and Ireland started from high levels of susceptibility to shocks on their cross-border investments, and these have come down significantly. The downward trend was mainly attributable to a lower volume of cross-border investments than to higher capital levels. In addition, the United Kingdom, France, Italy, Germany, Switzerland, and other coun tries (shown in the middle of Figure 13.5) also experienced downward trends.
  • In Greece, the susceptibility to network credit effects of cross-border investments increased over time.
  • The United States, Japan, Canada, Australia, India, and Turkey are some countries in the middle, where cross-border credit risks did not change significantly.

Individual Banking System’s Vulnerability to the Credit Shock

Sources: Bank for International Settlements; Bankscope; and authors’ estimates.

Note: Foreign claims of Finland are available after 2010. The graphs are placed in order of difference between 2008:Q4 and 2012:Q4 (ascending).

Interestingly, higher capital buffers seem to have largely contributed toward lower vulnerability to funding shocks, especially for two emerging economies for which there is published data. For India and Turkey and some larger countries, vulnerability to funding shocks came down since the crisis mainly due to higher capital levels. Simulations show that if capital (for all of the banking systems) was held constant at the end-of-2008 levels, then the vulnerability to bank funding flow reversals would have been going up. For the funding shock scenario, there could be two broad groups of countries— vulnerabilities trending down and unchanged (Figure 13.6):

  • The European countries in crisis— Ireland, Spain, Portugal, Greece— along with some others like the United Kingdom, India, and Turkey have been trending downward in their susceptibility to funding shocks. Among these, higher capital buffers seemed to have made a significant difference to India, Turkey, Canada, and the United Kingdom— making these countries more resilient to cross-border funding shocks.
  • In Austria, Germany, and Australia, cross-border funding vulnerability is largely unchanged.

Individual Banking System’s Vulnerability to the Funding Shock

Sources: Bank for International Settlements; Bankscope; and authors’ estimates.

Note: Foreign claims of Finland are available after 2010. The graphs are placed in order of difference between 2008:Q4 and 2012:Q4 (ascending).

There are also fewer propagators of network shocks than before. Comparing the global banking network at the end of 2008 to that at the end of 2012 (Figures 13.7 and 13.8), the number of “arrows” showing the direction of contagion have dropped. Back in 2008, the United States, the United Kingdom, France, and Germany were the main potential propagators (leading to at least 10 failures, or half the network) of credit shocks. France, Italy, and Germany were the main contributors to funding shocks. In 2012, the United States and the United Kingdom remained the key potential contributors of credit shocks. If the United States and the United Kingdom were to fail, there would be large ripple effects and failures in the rest of the world mainly from their borrowings from the rest of the world. Even though there are no longer major propagators of funding shocks, the United States, the United Kingdom, France, and Germany are still capable of having large impacts on at least two other economies due to funding shocks.2

Contagion to the Credit Shock and Funding Shock, 2008:Q4

Sources: Bank for International Settlements; Bankscope; and authors’ estimates.

Note: Blue sphere indicates the banking system that leads to more than 10 (that is, half of the number of countries in the dataset) induced banking failures. Arrows represent how shocks that lead to failure of the banking system are propagated. The figures are constructed with IMF data using the Excel add-in available at nodexl.com. A T = Austria; AU = Australia; BE = Belgium; CA = Canada; CH = Switzerland; DE = Germany; ES = Spain; FI = Finland; FR = France; GB = United Kingdom; GR = Greece; IE = Ireland; IN = India; IT = Italy; JP = Japan; NL = Netherlands; PT = Portugal; SE = Sweden; TR = Turkey; US = United States.

Contagion to the Credit Shock and Funding Shock, 2012:Q4

Sources: Bank for International Settlements; Bankscope; and authors’ estimates.

Note: Blue sphere indicates the banking system that leads to more than 10 (that is, half of the number of countries in the dataset) induced banking failures. Arrows represent how shocks that lead to failure of the banking system are propagated. The figures are constructed with our data using the Excel add-in available at nodexl.com. A T = Austria; AU = Australia; BE = Belgium; CA = Canada; CH = Switzerland; DE = Germany; ES = Spain; FI = Finland; FR = France; GB = United Kingdom; GR = Greece; IE = Ireland; IN = India; IT = Italy; JP = Japan; NL = Netherlands; PT = Portugal; SE = Sweden; TR = Turkey; US = United States.

Are the real effects of cross-border banking linkages well understood by macroeconomic forecasters? In what follows, the chapter tries to gauge whether greater vulnerability to cross-border banking network shocks are already taken into account in the GDP growth forecasts or whether there are major surprises. The answer is it depends upon whether the connections are on the assets or the liabilities side of the balance sheet.

5. What is the Output Cost of Vulnerability to Banking Interlinkages?

Extensive cross-border banking linkages bring both benefits and costs. Banking systems can share risk by diversifying their investments across borders so that there is no excessive reliance on good prospects at home. At the same time, banking systems have often relied on foreign funds to sponsor domestic credit growth when times are good or when banks are competing with other banks for market share in a specific loan segment. Both cross-border investments (asset growth) and funding (liabilities growth) carry the risk of reversal during a global crisis or a crisis from the other country. So, during good times, banking systems can grow and contribute to output growth. However, during stress in other countries, the cross-border credit and funding channels are conduits for bringing home crises from other countries and could have negative GDP growth surprises for the recipient banking system.

Cross-border banking linkages on the credit side do not seem to produce GDP surprises. A panel regression with country fixed effects is estimated to find out whether vulnerabilities to c ross-border credit and funding risks explain GDP growth surprises for the 20 countries in the sample (Table 13.2).3 The results show that cross-border credit linkages and the risks stemming from the linkages seem to be well understood by those making GDP forecasts. While the 2008–09 crisis had negative growth surprises on average for all countries, exposure to credit risk from other banking systems did not significantly make countries better of during normal times, nor did it infict damage beyond what was expected, during the crisis (Table 13.2, columns 1 and 2).

Table 13.2.Panel Regression with Country Fixed Effects(Dependent variable: GDP growth rate surprise; sample: 2005–12 [annual, fourth quarter])
λ = 0.5, ρ = 0.5(1)(2)(3)(4)(5)(6)
Crisis-3.03***

(0.29)
-3.64***

(0.63)
-3.16***

(0.28)
-1.28

(1.02)
-3.08***

(0.28)
-3.67***

(0.52)
Vul (credit)-10.02

(0.02)
0.01

(0.02)
Vul (credit)-1* Crisis0.03

(0.03)
Vul (funding)-10.05**

(0.02)
0.05**

(0.02)
Vul (funding)-1* Crisis-0.07*

(0.03)
Capital-1-8.12

(4.98)
-7.77

(4.96)
Capital-1* Crisis12.39*

(7.13)
Exposure-10.36

(0.27)
0.38

(0.29)
Exposure-1* Crisis-0.30

(0.27)
Observations140140140140140140
R-squared0.4940.4990.5130.5280.5080.521
Country pairs202020202020
Source: Authors.Note: Standard errors in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01. Capital = capital/GDP; Crisis = 2008–09; Dependent variable = growth rate surprise (actual GDP growth rate – GDP growth rate forecast) in percentage points; Exposure = (foreign claims + foreign liabilities)/GDP; Vul (.) = the vulnerability index from the network analysis.
Source: Authors.Note: Standard errors in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01. Capital = capital/GDP; Crisis = 2008–09; Dependent variable = growth rate surprise (actual GDP growth rate – GDP growth rate forecast) in percentage points; Exposure = (foreign claims + foreign liabilities)/GDP; Vul (.) = the vulnerability index from the network analysis.

By contrast, the real effects of possible funding reversals due to cross-border interlinkages during crises are not well understood. In good times, countries experience higher growth (surprises) by taking up cross-border funding risks, for instance by extending domestic credit funded from cross-border sources. The estimates (Table 13.2, columns 3 and 4) show that during normal (or noncrisis) times, every percentage point potential shortfall in capital levels contributes to a 0.05-percentage-point increase in GDP growth surprise. During crises, however, the benefits could reverse much more, leading to a 0.07-percentage-point decrease in GDP growth surprises over and above the average negative surprises. The same vulnerability reverses the good outcomes during crisis, although the Wald test on the sum of the coefficients on the funding vulnerability and the cross term is not always significantly different from zero.4,5

As is shown later in this chapter, a random-effects specification yields an even stronger result for the funding vulnerability. Every percentage point of potential capital depletion due to higher funding vulnerability increases surprises by 0.03 percentage point during normal times, and reduces surprises by 0.06 of a percentage point during crisis, and this effect is economically significant (the null hypothesis for the Wald test is rejected strongly).

Having higher capital buffers of the countries receiving the shocks helps during crises, and has no material impact on real growth surprises during normal times. To see if the measure on network vulnerabilities can be substituted by data on exposures and capital separately, a third set of regressions (Table 13.2, columns 5 and 6) was estimated.6 Results show that higher capital does not lead to lower growth surprises and higher exposures do not contribute to positive growth surprises, in general. However, during crises, having higher capital buffers helps to cushion the (negative) surprise impact.

Robustness

The results presented earlier in this chapter are generally robust to different assumptions on parameters for the network analysis and different specifications for the regressions.

  • Indices constructed with different lambda and rho: The movement of the indices is similar to the original indices if different parameters are used. The initial vulnerability measures are highly correlated (above 0.9) to the new indices constructed with different parameters. The trends in these indices are similar between various assumptions on the parameters for their construction: λ and ρ (Figures 13.9 and 13.10).
  • In the regression part, the findings regarding the funding and credit vulnerability indices are robust to various assumptions on the parameter values (λ and ρ) for the network analysis. The cross-product terms (crisis * vulnerability) are also still significant for most parameter values. Table 13.3 shows one such set of parameters.
  • Rerunning the regressions using random, instead of fixed, effects gives a stronger result on the funding risk (Table 13.4). As mentioned earlier in this chapter, higher funding vulnerability significantly exacerbates negative output surprises. In general, results of panel regressions with random effect are overall similar to the baseline result.
  • The result that higher capital buffers help cushion negative output surprises during crisis is robust to different model specifications and different data on capital from the IMF Financial Soundness Indicators database, where the data start in 2008, instead of Bankscope.

Credit Vulnerability Indices—Varying λ

(Weighted average of vulnerability with different parameters; global banking system, 2005–12)

Sources: Bank for International Settlements; Bankscope; and authors’ estimates.

Funding Vulnerability Indices—Varying ρ

(Weighted average of vulnerability with different parameters; global banking system, 2005–12)

Sources: Bank for International Settlements; Bankscope; and authors’ estimates.

Table 13.3.Robustness: Panel Regression with Country Fixed Effects (Dependent variable: GDP growth rate surprise; sample: 2005–12 [annual, fourth, quarter])
λ = 0.3, ρ = 0.3(1)(2)(3)(4)
Crisis-3.07***

(0.29)
-3.44***

(0.54)
-3.17***

(0.28)
-2.20***

(1.02)
Vul (credit)-10.04

(0.02)
0.03

(0.03)
Vul (credit)-1* Crisis0.03

(0.03)
Vul (funding)-10.11***

(0.04)
0.12**

(0.02)
Vul (funding)-1* Crisis140-0.07*

(0.03)
Observations140140140140
R-squared0.5000.5020.5190.535
Country-pairs20202020
Source: Authors.Note: Standard errors in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01. Capital = capital/GDP; Crisis = 2008–09; Dependent variable = growth rate surprise (actual GDP growth rate – GDP growth rate forecast) in percentage points; Exposure = (foreign claims + foreign liabilities)/GDP; Vul (.) = the vulnerability index from the network analysis.
Source: Authors.Note: Standard errors in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01. Capital = capital/GDP; Crisis = 2008–09; Dependent variable = growth rate surprise (actual GDP growth rate – GDP growth rate forecast) in percentage points; Exposure = (foreign claims + foreign liabilities)/GDP; Vul (.) = the vulnerability index from the network analysis.
Table 13.4.Robustness: Panel Regression with Random Effects (Dependent variable: GDP growth rate surprise; sample: 2005–12 [annual, fourth quarter])
λ = 0.3, ρ = 0.3(1)(2)(3)(4)
Crisis-3.04***

(0.28)
-3.65***

(0.61)
-3.00***

(0.28)
-0.48***

(0.96)
Vul (credit)-10.02

(0.01)
0.01

(0.02)
Vul (credit)-1* Crisis0.03

(0.03)
Vul (funding)-10.01

(0.01)
0.03*

(0.01)
Vul (funding)1 * Crisis140-0.09***

(0.03)
Observations140140140140
R-squared0.4530.4580.4410.473
Country-pairs20202020
Source: Authors.Note: Standard errors in parentheses. *p & 0.10, **p & 0.05, ***p & 0.01. Capital = capital/GDP; Crisis = 2008–09; Dependent variable = growth rate surprise (actual GDP growth rate – GDP growth rate forecast) in percentage points; Exposure = (foreign claims + foreign liabilities)/GDP; Vul (.) = the vulnerability index from the network analysis.
Source: Authors.Note: Standard errors in parentheses. *p & 0.10, **p & 0.05, ***p & 0.01. Capital = capital/GDP; Crisis = 2008–09; Dependent variable = growth rate surprise (actual GDP growth rate – GDP growth rate forecast) in percentage points; Exposure = (foreign claims + foreign liabilities)/GDP; Vul (.) = the vulnerability index from the network analysis.
6. Conclusions

To summarize, banking systems’ vulnerabilities from cross-border network linkages were found to have decreased in the last five years. For both asset-and liability-side vulnerabilities, on average for the global banking system, the potential for capital depletion arising from credit risks and funding risks has come down since the global financial crisis. The reduction is mainly due to lower exposures, but is also partly due to higher capital buffers around the world.

While the trend is similar for individual countries, the reason for the decline in vulnerabilities differs between countries and between credit and funding for particular countries. It was also found that, compared to 2008, the number of countries as core propagators of credit and funding shocks have dropped. The United Kingdom and the United States would still be the major propagators of credit shocks in 2012.

Funding risks have significant positive effects on growth surprises during normal times and significant negative effects on growth surprises during crisis times. Risks from cross-border borrowing have significant impacts on real growth surprise and these risks are higher than those from cross-border credit. Therefore, risks from cross-border borrowing need much more analysis and understanding than just looking at overall external funding volumes. In particular, taking on higher funding risks (by borrowing more from cross-border sources) generally exacerbates the negative output surprise during crisis. This finding is robust to different values of the parameters used to create the vulnerability indices and different specifications and estimation methods of the regression model.

Regardless of network effects, higher capital helps during a crisis, and it does not hurt to raise it during normal times. Higher capital buffers help mitigate negative GDP surprises during crisis, but the same buffers might not have a real impact during normal times. These findings give additional reasons for strengthening buffers during normal times, since it does not seem to have a significant impact on output surprises.

Future research could try to explain why funding risks appear to matter more than credit risks. One reason could be the transparency of credit links apparent with the published BIS data and a general understanding of the cross-border credit exposures of banks from certain countries. For instance, it is well known that the Spanish and Austrian banks have large credit exposures in Latin America and central and eastern Europe, respectively. However, there is less documentation about which countries Spanish and Austrian banks (and other sectors) borrow from. The BIS Consolidated Statistics do not provide liability-side information. As mentioned before, such information was only derived by making assumptions. Policymakers need to understand the specific vulnerabilities from funding linkages while making macroeconomic forecasts, and this chapter has made the case for the need to access better data.

References

    AlbertazziU. andM.Bottero. 2014. “Foreign Bank Lending: Evidence from the Global Financial Crisis.” Journal of International Economics92 Supplement 1 (0): S22S35.

    • Search Google Scholar
    • Export Citation

    CeruttiEugenio. 2013. “Banks’ Foreign Credit Exposures and Borrowers’ Rollover Risks Measurement, Evolution and Determinants.” IMF Working Paper 13/9International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Banks-Foreign-Credit-Exposures-and-Borrowers-Rollover-Risks-Measurement-Evolution-and-40235.

    • Search Google Scholar
    • Export Citation

    CeruttiEugenioStijnClaessens andPatrickMcGuire. 2011. “Systemic Risk in Global Banking: What Available Data Can Tell Us and What More Data Are Needed.” IMF Working Paper 11/222International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Systemic-Risks-in-Global-Banking-What-Available-Data-Can-Tell-Us-and-What-More-Data-Are-25259.

    • Search Google Scholar
    • Export Citation

    CetorelliN. andL. S.Goldberg. 2010. “Global Banks and International Shock Transmission: Evidence from the Crisis.” IMF Economic Review59 (1): 4176.

    • Search Google Scholar
    • Export Citation

    CihákMartinSoniaMunoz andRyanScuzzarella. 2011. “The Bright and the Dark Side of Cross-Border Banking Linkages.” IMF Working Paper 1/41International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/The-Bright-and-the-Dark-Side-of-Cross-Border-Banking-Linkages-25147.

    • Search Google Scholar
    • Export Citation

    De HaasR. andI.Van Lelyveld. 2014. “Multinational Banks and the Global Financial Crisis: Weathering the Perfect Storm?Journal of Money Credit and Banking46 (s1): 33364.

    • Search Google Scholar
    • Export Citation

    Espinosa-VegaMarco andJuanSolé. 2011. “Cross-Border Financial Surveillance: A Network Perspective.” Journal of Financial Economic Policy3 (3): 82205.

    • Search Google Scholar
    • Export Citation

    GiannettiM. andL.Laeven. 2012. “The Flight Home Effect: Evidence from the Syndicated Loan Market during Financial Crises .” Journal of Financial Economics104 (1): 2343.

    • Search Google Scholar
    • Export Citation

    International Monetary Fund (IMF). 2009. Global Financial Stability Report—Responding to the Financial Crisis and Measuring Systemic Risks Chapter 2. Washington, DCApril. https://www.imf.org/external/pubs/ft/gfsr/2009/01/.

    • Search Google Scholar
    • Export Citation

    International Monetary Fund (IMF). 2010. Global Financial Stability Report— Meeting New Challenges to Stability and Building a Safer System Chapter 2. Washington, DCApril. https://www.imf.org/en/Publications/GFSR/Issues/2016/12/31/Meeting-New-Challenges-to-Stability-and-Building-a-Safer-System.

    • Search Google Scholar
    • Export Citation

    International Monetary Fund (IMF). 2013. World Economic Outlook— Transitions and Tensions Chapter 3. Washington, DCOctober. https://www.imf.org/external/pubs/ft/weo/2013/02/.

    • Search Google Scholar
    • Export Citation

    Kalemli-OzcanSebnemEliasPapaioannou andFabrizioPerri. 2013. “Global Banks and Crisis Transmission.” Journal of International Economics89 (2): 495510.

    • Search Google Scholar
    • Export Citation

    KimKyunghun andSrobonaMitra. 2014. “Real and Financial Vulnerabilities from Cross-Border Banking Linkages.” IMF Working Paper 14/136International Monetary FundWashington, DC. https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Real-and-Financial-Vulnerabilities-from-Crossborder-Banking-Linkages-41792.

    • Search Google Scholar
    • Export Citation

    MinoiuCameliaChanhyunKangV.S.SubrahmanianandAna-maria Berea. 2015. “Does Financial Connectedness Predict Crises?” Quantitative Finance15 (4): 60724.

    • Search Google Scholar
    • Export Citation

    NierE.J.YangT.Yorulmazer andA.Alentorn. 2007. “Network Models and Financial Stability.” Journal of Economic Dynamics and Control31 (6): 203360.

    • Search Google Scholar
    • Export Citation

    PopovA. andG. F.Udell. 2012. “ Cross-Border Banking, Credit Access, and the Financial Crisis.” Journal of International Economics87 (1): 14761.

    • Search Google Scholar
    • Export Citation
1

Confidential bilateral data based on the BIS Locational Statistics, which was available for the third quarter of 2013, provides the breakdown by bank and nonbank exposures. On average, 60 percent of the cross-border claims of the BIS reporting banks resident or located in a certain country are on the banking sector; the average is higher for the G7 countries.

2

India and Turkey did not fall in the path of ripple effects through funding shocks from the United States, the United Kingdom, France, or Germany in 2012. Banking linkages do not help explain the turmoil in capital flows to India and Turkey experienced during the Federal Reserve tapering fears in the middle of 2013.

3

Growth surprise for a country is calculated by actual GDP growth rate minus the forecast of GDP growth rate from Consensus Forecasts.

4

The Wald test on the difference between normal and crisis times cannot reject the null hypothesis (H0: coefficient on funding vulnerability + coefficient on interaction with crisis dummy = 0).

5

A set of regressions with trade linkages was estimated but is not included in Table 13.2. The trade linkage is measured by (export to and import from the other 19 countries)/GDP. Trade linkages between these countries do not seem to matter for growth surprises during good times or bad times, nor do trade linkages change the outcomes for credit and funding vulnerabilities on growth surprises. This is because trade linkages are typically well documented and included in the dataset while making GDP growth forecasts.

6

Financial openness or exposure measured by aggregate statistics (foreign claims + foreign liabilities)/GDP is a standard regressor in a growth regression.

    Other Resources Citing This Publication