Channels of Crisis Transmission in the Global Banking Network Galina Hale (FRBSF) Tümer Kapan (Fannie Mae) Camelia Minoiu (IMF) *The views expressed herein are those of the authors and should not be attributed to the Federal Reserve System, Fannie Mae, the IMF, their Executive Boards, or their management. __ OECD __ non-OECD The Story in a Nutshell In 2010, Citibank NA had syndicated loan exposures vis-à-vis 105 banks in 94 countries. Motivation • Complexity of financial linkages has increased in recent times and may have contributed to the severity and reach of the global financial crisis • Complexity may play a role in how systemic banking crises are transmitted internationally – Ongoing efforts on banking regulation – Debate on what bank-specific “systemicness” is Question • Study the role of financial systemic complexity in the transmission of shocks worldwide: – Direct channels – Indirect channels • Study the impact of financial linkages on bank performance / profitability – Bank profitability matters because it predicts survival Aim • Disentangle the direct vs. indirect channels through which crises are transmitted globally: – Direct exposures • First degree connections – Indirect exposures • Higher degree connections – Relative position in the network • Centrality in the network Contribution • First paper to use bank-level lending data to 1. Construct global banking networks (GBN) 2. Compute bank-level measures of interconnectedness and 3. Directly relate interconnectedness to bank performance (2,000 interconnected banks are linked to their financials during 1997-2010) Hypotheses • Theory: interconnectedness carries both – Benefits: diversification, shock diffusion and – Risks: facilitates transmission of shocks/contagion • Banking linkages may play a different role during normal and crisis times – Normal times: portfolio diversification concerns, search for yield, advantageous market position – Crisis times: direct losses and contagion, but ability to spread losses through linkages and to leverage past connections to obtain funding during crunches Formally • Bank performance Y is affected by crises in its country C and performance of banks it is exposed to (directly or indirectly) network distance decay factor • Substituting for Yj direct indirect exposure exposure … adding network measures network Expanding: characteristics small and insignificant Data Construction • Use data on bank-to-bank lending from the international syndicated market for 1990-2010 from Dealogic’s Loan Analytics – Carefully clean up bank names, adjust for bank name changes, merges and acquisitions, etc. – Split total loan volume by bank (pro-rata) – Construct 2 GBNs • Merge with bank balance sheet data from Bankscope • Systemic banking crisis dates: Laeven and Valencia (2012) Example: Syndicated loan to a British investment bank Participating banks (15): Borrower: BayernLB; Bank of Montreal Investec Bank (UK) Ltd. (London); Bank of Tokyo-Mitsubishi UFJ Ltd; Commerzbank Industry: Private sector bank International Luxembourg SA; Dresdner Kleinwort Wasserstein; Signing date: March 28, 2006 HSH Nordbank AG (London); ING Bank NV; KBC; Lloyds TSB Bank plc; Deal type: Investment grade Mizuho Corporate Bank Ltd; Royal Bank of Scotland plc; SG Maturity: 3 years Corporate & Investment Banking; Standard Chartered Bank; Amount: GBP 445 million Sumitomo Mitsui Banking Corp Europe Ltd; Wachovia Bank NA Interest rate: LIBOR + 120bps Nationalities (7): Germany, UK, Japan, Luxembourg, Netherlands, Belgium, France Source: Loan Analytics 11 Exposure through syndicated lending to banks is about 37 percent of BIS loans to banks Exposures to banks in all countries 15 10 USD trillion 5 0 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 Syndicated loans BIS claims Global Banking Networks (GBN) 1. Network of current exposures EGBN – Based on bank-to-bank loan borrowing/lending exposures in the syndicated market: • Computed for every year t using information on outstanding loans (based on loan maturities) 2. Network of current and past relationships RGBN – Based on all current and past borrowing/lending relationships in the syndicated market: • Reflects entire history of financial transactions • Captures information flows and learning (Hale, 2012) • May capture linkages in other lines of business Both networks are directed; use only asset-based connectivity EGBN 2007 – subnetwork of 100 largest banks by 2007 assets Bank-to-bank Lending and Connections 60,000 500 55,000 total lending volume, current USD bn (right-axis) 450 # current relationships (EGBN) 50,000 # past relationships (RGBN) 400 45,000 350 40,000 300 35,000 30,000 250 25,000 200 20,000 150 15,000 100 10,000 50 5,000 0 0 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Bank-level Measures of Interconnectedness • Two concepts: – Distance: # of banks a bank has to go through to reach another bank (in the shortest possible way) – Proximity: 1/distance Bank-level Measures of Interconnectedness 1. Direct exposures – In USD or # of direct financial partners (distance=proximity=1) 2. Indirect exposure – Proximity to the banks from each country excluding directly linked banks (0<proximity<1) 3. Relative position in the network – Betweenness Centrality (“key intermediary”) Indicator for banks that link different groups of banks in the network – Closeness Centrality Average proximity to all the other banks in the network – Proximity to Network Center Proximity to the network’s most centrally-located bank Which Are the “Network Centers”? Year Bank Country Bank Country EGBN RGBN 1997 JP Morgan US JP Morgan US 1998 JP Morgan US LRP Germany 1999 BayernLB Germany WestLB Germany 2000 BayernLB Germany Citibank US 2001 BayernLB Germany Unicredit Germany 2002 BayernLB Germany HSBC UK 2003 WestLB Germany HSBC UK 2004 HSBC UK HSBC UK 2005 HSBC UK HSBC UK 2006 Santander Spain Santander Spain 2007 BBVA Spain BBVA Spain 2008 BBVA Spain BBVA Spain 2009 BBVA Spain BBVA Spain 2010 BBVA Spain BBVA Spain These banks have the highest closeness centrality in the binary and weighted EGBN and RGBN. What is a “Key Intermediary”? __ lenders __ borrowers In 2010, Arab Bank PLC (Jordan) had syndicated loan claims on 16 banks and liabilities vis-à-vis 29 banks . Empirical Approach Regression Set-Up Main Covariates • Panel regressions: • Direct and indirect – Dataset: 2,000 banks from 67 exposures: countries over 1997-2010 – Dependent variable: ROA, ROE – Computed vis-à-vis crisis and non-crisis countries • Controls: • $ exposures (EGBN) – Bank size (log-assets) • # current exposures (EGBN) – Capital (equity/assets) – Indicator for crisis in bank’s • # past exposures (RGBN) home country • Network position – Type of entity dummies – Closeness, betweenness, and – Specialization dummies proximity to network’s center – Bank country FE – Interacted with crisis in the – Year FE bank’s home country • St. errors clustered on bank Effect of Direct and Indirect Exposures on Bank Performance – Baseline (1) (2) (3) (4) L. Direct US$ current exposure to non-crisis countries 0.003 (0.019) L. Direct US$ current exposure to crisis countries -0.104*** (0.037) L. Direct current exposure to non-crisis countries -0.000 0.000 0.003* (0.001) (0.002) (0.001) L. Direct current exposure to crisis countries -0.012*** -0.009** -0.009* (0.004) (0.004) (0.005) L. Direct past exposure to non-crisis countries -0.001 -0.001 (0.001) (0.001) L. Direct past exposure to crisis countries -0.009** -0.009** (0.003) (0.003) L. Indirect proximity to non-crisis countries -0.028** (0.011) L. Indirect proximity to crisis countries 0.033 (0.038) Observations 9,129 9,129 9,129 9,129 R-squared 0.333 0.334 0.334 0.335 Effect of Direct and Indirect Exposures on Bank Performance – Robustness (1) (2) (3) (4) (5) (6) Average Cluster on Benchmark ROE Drop 1% Bank FE for country indirect L. Direct current exposure to non-crisis countries 0.003* 0.008 0.003* 0.001 0.001 0.003* (0.001) (0.022) (0.001) (0.002) (0.001) (0.002) L. Direct current exposure to crisis countries -0.009* -0.112* -0.008* -0.002 -0.008* -0.009* (0.005) (0.061) (0.004) (0.005) (0.004) (0.005) L. Direct past exposure to non-crisis countries -0.001 0.008 -0.001 0.002 -0.001 -0.001 (0.001) (0.014) (0.001) (0.002) (0.001) (0.002) L. Direct past exposure to crisis countries -0.009** -0.071 -0.010*** -0.009* -0.009*** -0.009** (0.003) (0.044) (0.003) (0.005) (0.003) (0.004) L. Indirect proximity to non-crisis countries -0.028** -0.195 -0.028** -0.029** -11.249* -0.028** (0.011) (0.132) (0.011) (0.014) (6.005) (0.011) L. Indirect proximity to crisis countries 0.033 0.768 0.035 0.046 12.519 0.033 (0.038) (0.496) (0.040) (0.047) (16.188) (0.049) Observations 9,129 9,128 8,239 9,129 8,322 9,129 R-squared 0.335 0.203 0.328 0.557 0.330 0.335 Effect of Relative Position in the Network on Bank Performance - Baseline (1) (2) (3) (4) L. Key intermediary 0.057 0.149* (0.074) (0.077) L. Key intermediary * Crisis in home country -0.286* -0.538*** (0.173) (0.185) L. Closeness centrality -0.848** -1.920*** (0.391) (0.649) L. Closeness centrality * Crisis in home country 1.439* 6.014*** (0.783) (1.352) L. Closeness to network center 0.009 0.142** (0.039) (0.065) L. Closeness to network center * Crisis in home country -0.227** -0.627*** (0.098) (0.155) Benchmark regressors included? yes yes yes yes Observations 8,342 8,342 8,342 8,342 R-squared 0.331 0.331 0.331 0.335 Conclusions: The “dark side” of being connected • Direct current & past exposures to banks in crisis countries reduce profitability (effect is small)