A Contagion Through Exposure to Foreign Banks During the Global Financial Crisis

Although the global financial crisis of 2008 took root in the advanced countries, its shocks spread through the emerging economies, reflecting the increasingly interconnected global financial system. This paper develops an empirical methodology to test the contagion effect at the country level using bilateral data on bank claims between countries. It measures the direct and indirect exposures of emerging economies to crisis countries and tests whether these matter for capital outflows from emerging economies. The paper measures these exposures to the crisis-affected countries by using bilateral foreign claims sourced from Bank for International Settlements (i) consolidated banking statistics foreign claims on immediate counterparty and ultimate risk bases and (ii) locational banking statistics cross-border total claims. Findings show that emerging market economies more exposed directly or indirectly to banks in the crisis-affected countries suffered more capital outflows during the global financial crisis.


Introduction
Although the global financial crisis of 2008 took root in the advanced countries, its shocks spread through the emerging economies, reflecting the increasingly interconnected global financial system. The deterioration of subprime loans that seeded the crisis hit banks in advanced countries directly, forcing them to retreat and curtail their international exposures in search for liquidity. The market for short-term loans dried up and some of these troubled banks had to withdraw their funds from emerging economies. This in turn, led to a liquidity crunch among banks in emerging economies that had borrowed short-term funds from banks in advanced countries to provide long-term loans to domestic borrowers. These spillovers tended to be disproportionately high when the troubled banks in advanced countries were larger and more interconnected, that is, they are systemically more important banks.
Clearly, the growing global financial network and interconnectedness can amplify and transmit a shock from one bank to another, leading to systemic crisis. Providing micro-level evidence of such transmission is very difficult, however, due to a lack of data representing the comprehensive banking network and individual banks' international exposures. Morrison et al. (2016) were able to identify bank credit default swap (CDS) returns attributable to counterparty losses through the network of CDS transactions between banks.
Their findings show that information about counterparty losses is transmitted to a bank's own CDS spread. That is, whenever the counterparties from which a bank has purchased default protection experience losses, the likelihood of endorsing the default protection decreases and the CDS spread of the bank increases. Hale, Kapan, and Minoiu (2016) also find that profitability and loan supply decline in banks with direct or indirect exposures to countries experiencing systemic banking crises. This paper develops an empirical methodology to test the contagion effect at the country level using bilateral data on bank claims between countries. Our goal is to empirically measure direct and indirect exposures of emerging economies to crisis countries and test whether these exposures matter for capital outflows from emerging economies. The country level bilateral bank claims data are collected from Bank for International Settlement (BIS) consolidated banking statistics and locational banking statistics. １ １ BIS compiles and publishes two sets of statistics on banks' international positions. Consolidated banking statistics measure banks' country risk exposures by capturing the worldwide consolidated claims of internationally active banks headquartered in BIS reporting countries. Locational banking statistics provide information about the currency composition of banks' balance sheets and the geographical breakdown of their counterparties by capturing outstanding claims and liabilities of banks located in BIS reporting countries, including intragroup positions between offices within the same banking group.
While financial interconnectedness can arise from both the asset and the liability sides of banks' balance sheets, earlier studies on banking crises focused more on the asset side of interconnectedness for financial contagion and spillovers. For example, the first-generation models of banking crises considered how an economic downturn or recession would undermine corporate borrowers' ability to service their debts and impair bank assets, setting off bank runs (Mishkin 1978). Another set of studies paid attention to a lending boom and increased financial leverage, often followed by a subsequent burst and collapse in asset prices, leading banks to scale back their lending (Demirgüç-Kunt and Detragiache 1998). Allen and Babus (2009) also reviewed a group of papers that used network theory to explain financial contagion. Not only do bank failures spread directly through mutual claims they have on one another, but they can also spread indirectly through forced sales of assets by some banks that depress the market price, inducing further distress in other institutions.
However, recent studies have noted the risk of contagion through bank exposures on the liabilities side. Shin (2009) reflects on the Northern Rock bank run in the United Kingdom, and notes that when risk constraints took effect, lenders cut back their exposures in response to the crisis. He points to potential disruption in financial systems caused by a sharp pullback in leverage as creditors adjust their risk exposure. He notes that creditors' deleveraging actions, while prudent from their risk management perspective, would look like a "run" from Northern Rock's viewpoint. Čihák, Muñoz, and Scuzzarella (2011) further note that it is important to determine whether the cross-border interlinkages are stemming primarily from banks' asset or liabilities sides. Using measures that differentiate the types of interconnectedness, they find that the impact of changes in interconnectedness on banking system fragility are more significant for liability-side (upstream) interconnectedness than for asset-side (downstream) interconnectedness. That is, financial turmoil originated in creditor countries and moving upstream through borrowing countries' funding channels is found to have more detrimental economic impact on the borrowing countries than financial turmoil originated in borrowing countries and moving downstream to the creditors.
Our paper empirically investigates how shocks are transmitted through banks' exposures on the liability side. We find that the more emerging economies' liability sides were exposed directly or indirectly to crisis countries during the global financial crisis, the higher the rate of capital outflows they suffered. During the global financial crisis, as banks in advanced countries experienced liquidity and risk constraints, they reversed their lending positions against banks in emerging economies. Our findings suggest that deleveraging of the banks in advanced economies then triggered a run on banks and other entities in emerging economies. ２ The reminder of the paper is organized as follows. In the next section, we explain the data used in our empirical analyses. In section 3, we lay out our empirical framework of calculating direct and indirect exposures of emerging economies to crisis countries. In section 4, we report and discuss the main empirical findings of the paper. Section 5 concludes.

Data
Bilateral data on cross-border liability positions are collected from BIS consolidated banking statistics and locational banking statistics. The consolidated banking statistics provide consolidated claims of internationally active banks headquartered in 30 BIS reporting countries against 223 counterparty countries. ３ In these statistics, the claims of banks' foreign affiliates are included but intragroup positions are excluded, similar to the consolidation approach followed by financial regulatory supervisors. ４ The statistics also report the transfer of credit risk from the immediate counterparty to the country of ultimate risk (where the guarantor of a claim resides). Locational banking statistics report the outstanding claims of banks located in 43 BIS reporting countries. ５ The important difference between consolidated and locational banking statistics is that the latter include intragroup positions between offices of the same banking group if they are in different countries.
Since the number of the BIS reporting countries is limited, that is, there are other claims of banks with controlling parents located outside the BIS reporting countries, the sum of all claims of these reporting countries against a counterparty country would not be equal to the sum of all liabilities held by the counterparty country. However, since the BIS reporting countries include most countries active in international bank loans, actual total foreign claims on a counterpart country are not expected to deviate too much from the sum of the claims of the banks of these reporting countries only. ６ ２ See, among others, Shin (2009), which emphasizes the liability side of the balance sheet and its implications for a bank run. ３ In the consolidated banking statistics, claims refer to outstanding loans and holdings of securities by reporting banks. Appendix The locational banking statistics report the claims of all banks resident in 43 BIS reporting countries. The claims are broken down by instrument, currency, sector, country of residence of counterparty, and nationality of reporting banks. ７ Since the organization principle underlying the reporting requirement of the local banking statistics is the location of the banking office, the statistics include international transactions of a bank with any of its own affiliates outside the reporting country, consistent with balance of payments and external debt methodology. The claims cover deposits and balances placed with banks, loans, and advances to banks and nonbanks and holdings of securities and participations.
In the locational banking statistics, banks' total international claims are decomposed into three categories: (i) loans (ii) debt securities and (iii) other assets. The last category (iii) includes equity shares, participations, derivative instruments, and working capital supplied by head offices to their branches abroad. The sectoral breakdown of banks' total international claims on (i) banks and (ii) nonbanks is also available. In fact, the locational banking statistics report total claims and claims on nonbanks, and claims on banks are calculated by subtracting the latter from the former. In addition, loans are again disaggregated into those to the banking and to the nonbanking sectors.
Consolidated banking statistics report the claims at an aggregate level compiled in two ways: by immediate counterparty and by ultimate risk. ８ The immediate counterparty is the entity in the counterparty country with which the bank in the reporting country counteracts directly. Ultimate risk is the counterparty country to which credit exposure transfers through credit risk mitigants such as collateral, guarantees, and credit protection. For example, suppose a bank of the United States (US) extends a loan to a company in a counterparty, country A, and the loan is guaranteed by a bank of another country, country B. Then, based on an immediate counterparty basis, country A is reported as the counterparty country of the US because the US bank reports the loan as a claim on the company in country A. On an ultimate risk basis, however, country B is reported as the counterparty country of the US because, if the company in country A defaults, then ultimately the claims will be made to the bank in country B that guarantees the loan.
On an immediate counterparty basis, foreign claims by the nationality of reporting banks are decomposed into (i) international claims and (ii) local positions in local currency. counterparties in Argentina is $44,902 (in millions) while total foreign claims including those of non-reporting countries is $45,494 (in millions). ７ See BIS (2012) for a detailed description of the nature of the locational banking statistics, which this paper mostly follows. ８ See the explanation in BIS (2016) for the difference between immediate country and ultimate risk bases.

5
The international claims of a reporting country include both cross-border claims of a bank headquartered in the reporting country and local claims in foreign currencies provided by its foreign affiliates. International claims are again disaggregated using two different approaches: (i) by remaining maturities in which it is divided into (a) up to and including 1 year, (b) over 1 year and up to 2 years, and (c) over 2 years; and (ii) by sectors of the counterparty country, which are (a) banks, (b) nonbank financial sector, (c) other private sector and (d) official sector.
Local positions in local currency refer to credits in local currency provided by foreign affiliates of the banks headquartered in the reporting country.
On an ultimate risk basis, foreign claims by nationality of reporting banks are disaggregated in two different ways: (i) by the type of position, which are (a) cross-border claims and (b) local claims; and (ii) by the four sectors the international claims by immediate counterparty are divided into (as indicated in the previous paragraph). The main difference in how the two reporting bases are disaggregated is that, for the ultimate risk basis, local claims denominated in local currencies and international claims are not reported separately, no maturity breakdown is available, and foreign claims, not international claims, are disaggregated into four different sectors. ９

Methodology
In this section, we develop two measures-direct and indirect-of emerging economies' exposures to crisis countries' banks through a network of bank claims. １０ Direct exposure of foreign claims on an emerging economy i at time t to banks in crisis countries, denoted by , is measured by the sum of shares of foreign claims held by all countries that experience crises: where N is the number of reporting countries, stands for the foreign claims held by reporting country k on counterparty country i at time t and hence ℎ , is the share of foreign claims held by reporting country j on country i at time t. Since is a set of reporting countries that experience crises at time t, the direct exposure is defined as the sum of shares of foreign claims across all reporting countries that experience crisis. While it would be more desirable to measure foreign claims and the degree of liquidity shocks at a bank level, this kind of data is not available. Instead we assume that all banks in a reporting country that experiences a crisis are faced with the same degree of liquidity problems.
The direct measure captures the idea that a country is exposed to the crisis-affected countries through its borrowing from them as the crisis-affected countries facing liquidity and risk constraints would withdraw funds or reverse their lending. The more a country borrowed from the crisis-affected countries, the more vulnerable is the country in concern to a reversal in capital flows. In Figure 1, which shows how the direct exposure is measured, each emerging economy is affected directly from crisis countries only (red circles), and the measure is calculated by summing up the red arrows that denote shares of foreign claims by reporting countries for each emerging economy.

<<Figure 1: Direct Exposures>>
However, the direct measure alone cannot fully capture a country's exposure to the crisis-affected countries, because it does not consider the country's exposure to all other countries that are not directly hit by the crisis but exposed to the crisis-affected countries and therefore face liquidity problems indirectly. For example, suppose country A itself does not experience a crisis, but it is exposed directly to countries that do, i.e., it is affected through the contagion effect measured by the direct exposure to crisis countries. Then an emerging country B that borrows funds from banks in country A can be indirectly affected from crisis countries through country A.
The above arguments imply that we can also define an indirect exposure of foreign claims of an emerging economy i at time t, , , as follows: Note that the indirect exposure of foreign claims is the weighted average of direct exposures faced by all reporting countries, with their shares of claims used as the weights. In Figure 2, showing how indirect exposure is measured, indirect exposure is indicated by the grey arrows.
Note that the grey arrow comes even from a non-crisis country (a blue circle) that is affected from other crisis countries (red arrows). Utilizing these disaggregate data, we define direct and indirect exposures of the banking sector, , and , , based on claims on the banking sector in the counterparty country: where ℎ , is the share of foreign claims held by country j on the banking sector in country i at time t. These exposure measures based on the banking sector are particularly interesting since the liabilities of the banking sector play a crucial role in transmitting shocks.
For example, Hahm, Shin, and Shin (2013), defining noncore liabilities of the banking sector as consisting mostly of banks' borrowings from foreign countries, showed that a large stock of noncore liabilities indicates the erosion of risk premiums and hence of vulnerability to a crisis.
We also define direct and indirect exposures of short-term maturities, , and , , using data on claims of maturities with less than 1 year on the counterparty country as follows: where ℎ , is the share of short-term claims held by country j on country i at time t. As explained, short-term claims are available only for consolidated banking statistics international borrowings on immediate counterparty basis. Since long-term claims are not easily withdrawn even by troubled banks, we expect that a bank-run type of sudden withdrawals of claims from emerging economies is more likely to take place in short-term borrowings by stopping rollovers.
We hypothesize that countries more exposed, both directly and indirectly, to banks in crisis countries suffered from more capital outflows during the global financial crisis. To test this hypothesis, we measure the rate of capital outflows from country i during the global financial crisis, , as follows:

Empirical Findings
For country i, we choose 65 emerging economies from the set of counterparty countries １１ .
The list of emerging economies is identical to that in Park, Ramayandi, and Shin (2016) that also follows Eichengreen and Gupta (2015) and Lim, Mohapatra, and Stocker (2014).
<<Table 1. Summary Statistics>> Table 1 shows summary statistics of the variables used in the regressions below.
Measures of direct and indirect exposures are calculated for 57-62 emerging economies depending on the choice of consolidated and locational banking statistics and sectors.
Generally, the measure of direct exposures is higher than that of indirect exposure, except for short-term debts, for which they are almost the same. We also report statistics for various bilateral claims. Note that, according to the consolidated banking statistics foreign claims on an ultimate basis and locational banking statistics cross-border total claims, the volume in the banking sector is about half the size of that in the entire sector. If you restrict to cross-border loans in the locational banking statistics, the volume of claims in the banking sector is about 80% of that in the entire sector.  Measures of direct and indirect exposures are calculated by using bilateral foreign claims in three data sources: consolidated banking statistics foreign claims on immediate counterparty (Table 2A) and ultimate risk bases (Table 2B), and locational banking statistics cross-border total claims (Table 2C) １２ We also measured direct and indirect exposures in Q1 2008, and the main results did not change. １３ See Eichengreen and Gupta (2015) and Park, Ramayandi, and Shin (2016) for the motivation for including these as explanatory variables. ultimate risk basis (Table 3A), and locational banking statistics cross-border total claims (Table   3B) and cross-border loans (Table 3C). In all three, columns (1) We also find that the coefficient of indirect exposure of the short-term maturity is not statistically significant in column (4). As for the banking sector, therefore, the direct exposure of the short-term borrowings seems to play a more crucial role in explaining the degree of vulnerability of the emerging economies.

Conclusion
In this paper, we investigated and tested financial contagion from advanced to emerging market economies through the global banking network. We computed measures of a country's direct and indirect exposures to the crisis-affected countries by using bilateral foreign claims sourced from (i) consolidated banking statistics foreign claims on immediate counterparty and ultimate risk bases and (ii) locational banking statistics cross-border total claims. Our findings show that emerging market economies that were more exposed, both directly and/or indirectly, to banks in the crisis-affected countries suffered from more capital outflows during the global financial crisis.
Empirical evidence suggests that a country's direct and indirect exposures to the crisis-affected countries can affect the size of capital outflows from the country during the crisis.
Overall, the degree of the country's direct exposure through the banking sector to the crisisaffected countries is the most important for capital outflows during the crisis. However, using locational banking statistics, indirect exposure becomes significant. This may reflect the added information value of the geographical breakdown of banks' counterparty exposures, including intragroup positions between offices within the same banking group, which would unlikely be captured by sovereign credit rating assessment.
Our findings also suggest that the global banking network of aggregate cross-border lending can be an avenue for global liquidity crunch to spread financial shock around the world.
This liquidity issue of creditor banks can be particularly problematic for emerging market economies that rely heavily on foreign borrowing. The findings also support the argument in recent financial contagion studies, such as Shin (2009), that countries that are not directly affected by a crisis can also experience financial disruption due to deleveraging actions by creditors in crisis-affected countries. That is, increased global financial network leaves no safe havens from a financial crisis. Also, as in Čihák, Muñoz, and Scuzzarella (2011), our findings show that financial turmoil originated in creditor countries can spread to borrowing countries through their funding channels.