## 1. Sample of Global Systemically Important Banks and Insurers

## Source details

**Canonical URL:** [1. Sample of Global Systemically Important Banks and Insurers](https://www.imf.org/-/media/files/publications/wp/2017/wp17210.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17210.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17210.pdf.json)

---

### Introduction and motivation
- Global Financial Crisis of 2007–2009 highlighted role of global systemically important banks and insurers in cross-border shock transmission.
- Regulatory lists maintained by the Financial Stability Board (FSB) since 2011; paper examines GSIBs and GSIIs as identified by FSB lists published in 2016, 2015 and 2014.

### Scope, data, and methodology
- Sample span: October 2007 to August 2016.
- Entities: 29 GSIBs and 10 GSIIs (39 financial institutions) located in the United States, Europe, and Asia.
- Market data:
  - Publicly-available daily equity return data (two-day log returns for time-zone adjustment).
  - Intra-day equity volatility (Parkinson and Diebold–Yilmaz methods).
- Data sources: Datastream for equity prices; assets data from SNL and Moody’s where referenced; CDS from Datastream for robustness checks (CDS sample: December 2007 to August 2016).
- Treatment of missing data: remove a day if more than half of entities missing; interpolate remaining missing observations.
- Primary empirical approach: Diebold and Yilmaz (2014, 2015) spillover framework (VAR-based generalized variance decomposition mapped to weighted directed networks).
- High-dimensional VAR estimation: elastic net estimator (Zou and Hastie, 2005).
- Complementary tools and measures referenced:
  - SRISK (Acharya et al. (2010, 2012))
  - Economic policy uncertainty (Baker, Bloom, and Davis (2016))
  - CoVaR (Adrian and Brunnermeier (2016)), SyRIn (Segoviano et al. (2017, forthcoming))
  - Linear and quantile regressions, panel regressions, scenario analysis.

### Main contributions
- Estimate market-based interlinkages among GSIBs and GSIIs, including insurance sector role.
- Compare Diebold–Yilmaz market-connectedness with balance-sheet SRISK.
- Assess explanatory power of risk indicators (GPU, VIX, short and long bond yields) for market connectedness.
- Analyze bank balance sheet drivers (profitability, asset quality) of bank-specific return connectedness.

### Stylized facts: size, returns, and volatilities
- By assets: largest GSIBs located in Asia, followed by Europe and the United States.
  - Example: ICBC and China Construction Bank combined asset value of US$6.5 trillion.
- Insurer vs bank assets:
  - Insurer asset size lower on average, but AXA, Allianz and Metlife comparable to Goldman Sachs, Morgan Stanley, and Credit Suisse.
- Average daily returns (sample average two-day log returns): European banks, Asian insurers, and U.S. insurers have the lowest (most negative) daily returns over October 2007–August 2016; overall sample equity returns negative despite recovery episodes.
- Intra-day volatility:
  - Daily variance estimator formula and annualized volatility expressed as in source (log-volatility used to approximate normality).
  - Empirical averages: European banks and U.S. insurers highest intra-day volatility on average; U.S. banks next; Asian banks lower. Volatilities right-skewed.

### Empirical findings — dynamic connectedness and crisis episodes
- Rolling estimation parameters:
  - Rolling window width = 200 days; predictive horizon for variance decomposition = 12 days (also robustness with 150 days and 24-day horizon examined).
- Total return and volatility connectedness rise during crises.
- Global Financial Crisis (peak October 2008, shortly after Lehman Brothers filing on September 15, 2008):
  - Returns total connectedness: 92.5 percent
  - Volatility total connectedness: 87.5 percent
- Greek sovereign debt crisis (September 2010 peak; stayed high until February 2011):
  - Returns: 89 percent (September 2010)
  - Volatility: 79 percent (September 2010)
  - Declined to Returns: 84.4 percent (summer 2011); Volatility: 70.8 percent (summer 2011) as bailout restored confidence.
- Renewed European tensions episode:
  - Returns: 93.2 percent (beginning of episode)
  - Volatility: 83.2 percent (beginning of episode)
- October 2012 levels after European policy measures:
  - Returns connectedness: 88.4 percent
  - Volatility connectedness: 70 percent
- Volatility connectedness rose around U.S. “fiscal cliff” (January 2013) and debt ceiling debate (October 2013), peaking around 75 percent at beginning of 2013.
- From July 2014, connectedness increased, reaching returns 87.6 percent and volatility 77 percent at beginning of 2015.
- Summer 2015 Chinese stock market disturbances and beginning-2016 European bank selloff drove renewed spikes; connectedness since January 2016 rose sharply to crisis-like levels.

### Regional, pairwise, and net directional connectedness
- Regional clustering:
  - Strongest pairwise connectedness concentrated within Asian, European, and U.S. clusters (top 15 percent linkages within regions).
- China:
  - Chinese GSIBs each explain around 19 percent of equity returns of other Chinese institutions.
  - Chinese GSIIs’ equity returns explain about 10 percent of Chinese GSIBs’ returns and vice versa.
- United States:
  - Two large U.S. GSIIs each explain about 7 percent of the other’s equity returns.
  - Among U.S. GSIBs: strongest connectedness among the four largest commercial banks at 6.2 percent on average; between the two large investment banks at 6.4 percent.
  - Highest bank–insurer connectedness among U.S. GSIB–GSII links at 5.3 percent.
- Europe:
  - Two Spanish banks explain about 6.4 percent of each other’s equity returns.
  - Large French GSIBs connectedness at 4.9 percent.
  - Two British insurers connectedness at 4.9 percent.
- Net directional connectedness (system-wide averages, October 2007–August 2016):
  - U.S. banks and insurers are the most important net sources of directional connectedness for returns and volatilities.
  - Asian GSIBs and GSIIs tend to be net recipients.
  - European GSIBs and GSIIs show moderate average net connectedness with cross-country variation.
- Dynamics:
  - U.S. banks and insurers dominant net contributors to return connectedness for majority of period, with elevated contributions during 2007–2009 GFC.
  - Volatility connectedness: European institutions overtook U.S. counterparts during mid-2009 to end-2013 and beginning of 2016; by end-sample (August 2016) European banks and insurers key drivers of volatility connectedness, with U.S. and Asian institutions as net receivers.
- Sectoral roles:
  - GSIBs from United States, Germany, Spain, and France are specific sources of outward returns and volatility connectedness.
  - GSIIs from United States, France, and Germany are specific outward sources.
  - GSIBs from China and Japan tend to be recipients based on market data.
  - Largest GSIBs often recipients; largest GSIIs tend to be sources.

### Robustness checks and comparison with alternative measures
- Robustness approaches:
  - Condition returns on global and country-specific returns and bond yields (filtered residuals).
  - Alternative market prices: CDS spreads (log-differences); CDS coverage limited—several institutions excluded due to lack of coverage.
  - Vary estimation window (150 vs 200 days), lag order (2 vs 3), and forecast horizon (12 vs 24 days).
- Main robustness conclusion: total connectedness index largely unchanged; elevated interconnectedness apparent during GFC, European Sovereign Debt Crisis, and recent periods.
- Comparison with SRISK:
  - SRISK: expected capital shortfall measure using LRMES with six-month crisis threshold of 40 percent; firm beta (β) from CAPM; prudential capital requirement k = 8 percent for Africa/Asia/Americas, k = 5.5 percent for Europe.
  - Broad trends in market-based connectedness accord closely with SRISK (bi-annual plots across 11 domiciliary countries).

### Common factors and drivers of global connectedness
- Market risk indicators examined: MOVE, VIX, U.S. Dollar Index (DXY), GPU (GDP-weighted PPP-adjusted average for 16 countries), US3 (3-month yields), US10 (10-year yields).
- Pair-wise correlations (Appendix Table 1; sample July 2008–August 2016):
  - GPU: 0.57
  - SLOPE: -0.40
  - US10: -0.38
  - VIX: 0.29
  - US3: 0.22
  - MOVE: 0.21
  - DXY: 0.15
- Observations:
  - GPU most strongly correlated with global connectedness in absolute terms.
  - Co-movement between global connectedness and MOVE and VIX not consistently strong since mid-2014.
  - Upswing in connectedness since late 2014 accompanied by steady appreciation of the U.S. dollar (DXY).
  - 10-year yields (US10) display countercyclical relationship with connectedness; principal component of 10-year yields across 11 countries moves in line with US10.
- Dynamic impulse responses (Box 1):
  - Bivariate VARs (GPU–GC and US10–GC) with VIX as exogenous control; samples:
    - Full: July 2008–August 2016
    - Subsample A: July 2008–August 2012
    - Subsample B: August 2012–August 2016
  - Connectedness responds positively to a one standard deviation shock to GPU and negatively to a shock to US10 across samples; 24-month horizon; 95 percent confidence bands via 20,000 Monte Carlo simulations.

### Predictive regression and quantile regression (Box 1 results)
- Predictive specification (C1): GC_t = alpha + beta·GPU_t + sigma·US10_t + phi·VIX_t + theta·GC_{t-1} + epsilon_t, with epsilon_t following GARCH(1,1).
- Linear regression results (July 2008 – August 2016):
  - alpha = 21.65** (4.50)
  - beta (GPU) = 0.01** (0.00)
  - sigma (US10) = -0.52** (0.26)
  - phi (VIX) = 0.04** (0.02)
  - theta (lagged GC) = 0.74** (0.05)
  - a0 = 1.01 (0.80)
  - a1 = 0.58 (0.39)
  - b1 = 0.33 (0.61)
  - R-squared = 0.89
  - (**) denotes significance at the 5 percent level.
- Quantile regression (τ = 0.1 to 0.9):
  - US10 more influential on higher percentiles of GC (τ ≥ 0.60).
  - GPU exhibits uniform impact across percentiles.
  - Examples:
    - alpha at τ = 0.1: 28.83** (7.35)
    - sigma (US10) at τ = 0.9: -1.08** (0.18)
    - phi (VIX) at τ = 0.9: 0.09** (0.01)
    - theta ranges: 0.64** (0.08) at τ = 0.1; 0.76** (0.03) at τ = 0.9.

### Disaggregated predictive regressions (To- and From-degrees) — representative coefficients (Table 3)
- Estimation sample: July 2008–August 2016; GARCH(1,1) effects important at disaggregated level.
- To-degree (GSIBs) examples:
  - U.S. alpha: 46.90** (6.57)
  - Europe alpha: 18.81** (5.06)
  - Asia alpha: 3.46 (3.30)
  - Sigma (US10) U.S.: -3.95** (0.93)
  - Theta (lagged) U.S.: 0.58** (0.04); Europe: 0.82** (0.05); Asia: 0.73** (0.06)
- From-degree (GSIBs) examples:
  - U.S. alpha: 27.60** (4.30)
  - Europe alpha: 21.77** (3.29)
  - Asia alpha: 28.19** (4.12)
  - Sigma (US10) U.S.: -0.98** (0.17); Europe: -0.84** (0.17); Asia: -1.54** (0.48)
  - Theta (lagged) examples: U.S. 0.71** (0.04); Europe 0.76** (0.04); Asia 0.66** (0.05)
- Heterogeneity: coefficients and GARCH parameters vary markedly by region and sector (GSIB vs GSII), indicating idiosyncratic dynamics alongside systemic drivers GPU and US10.

### Bank-specific panel regression results (Arellano-Bover/Blundell-Bond; Table 4)
- Sample span: 2008 to 2015; Bank_id = 29; Observations = 195 per regression.
- Dependent variables: To_Connectedness and From_Connectedness (lagged dependent variable included).
- Key coefficients (standard errors in parentheses):
  - To_Connectedness:
    - Lagged dependent variable: 0.467*** (0.105)
    - ROAA: -7.905** (3.466)
    - NPL ratio: 1.077 (1.189)
    - Total assets (ln): -3.148 (5.392)
    - Global uncertainty: 0.0624 (0.0761)
    - US10-year interest rate: 0.0173 (2.540)
  - From_Connectedness:
    - Lagged dependent variable: 0.229** (0.0928)
    - ROAA: -4.807*** (1.770)
    - NPL ratio: 0.906* (0.485)
    - Total assets (ln): -1.666 (1.033)
    - Global uncertainty: 0.130*** (0.0267)
    - US10-year interest rate: 1.948* (1.084)
    - Constant: 65.51*** (18.08)
- Interpretation:
  - Lower profitability (ROAA) associated with higher outward (to-) and inward (from-) connectedness.
  - Higher NPL ratio associated with higher inward connectedness.
  - Elevated global uncertainty and higher US10 associated with increased inward connectedness on average.

### Projections and counterfactuals (Box 2 and out-of-sample analysis)
- Out-of-sample prediction origin: August 2016 (model estimated July 2008–April 2016 for predictions April 2016–August 2016).
- Observed conditioning-path (August 2016 → January 2017):
  - US10 increased from 1.55 percent (August 2016) to 2.43 percent (January 2017).
  - GPU increased from 175.73 to 270.15 (reported as a 70 percent increase).
  - VIX remained close to six-month average.
  - Predicted GC oscillated slightly around prediction-origin level.
- Counterfactual 1:
  - US10 rise with GPU flat-lined at ~166 (12-month prior average) would set GC on a downward trajectory.
- Counterfactual 2:
  - To offset the observed 70 percent GPU increase (175.73 → 270.15) and generate the same downward GC trajectory, US10 would need to rise extremely steeply (described as implausibly large in source).
- Box 2 projection (U.S. growth scenario; conditioning on historical September 2002–December 2004 patterns):
  - US10 during September 2002–December 2004 rose from 3.87 percent to 4.23 percent.
  - Conditioning on US10 trajectory adapted to avoid level shift and median GPU path suggests a fall in global connectedness to early 2015 level over approximately 2.5 years.

### Interpretations and policy implications
- Market-based connectedness metrics (Diebold–Yilmaz) align broadly with balance-sheet SRISK trends and provide cross-jurisdictional insight when balance-sheet cross-border exposures are confidential.
- Key empirical conclusions:
  - GSIBs in United States, Germany, Spain, and France are important sources of directional connectedness.
  - GSIIs from United States, France, and Germany influence other GSIIs’ returns and volatilities.
  - Strong regional clusters and intense cross-sectoral within-region connectedness (bank–insurer) exist.
  - Total system connectedness rises during financial stress episodes.
  - Higher global economic policy uncertainty and lower U.S. 10-year bond yields are associated with increased market connectedness.
  - Bank-specific factors (profitability, asset quality) significantly affect return connectedness.
- Policy recommendations:
  - Intensify risk management and supervision of GSIBs in prominent domiciles (United States, Germany, Spain, France); monitor cross-border exposures.
  - Intensify supervision of prominent GSIIs (France, United States, Germany) with attention to inward and outward spillovers.
  - Monitor cross-sectoral exposures and pursue comprehensive group-wide supervision across banking and insurance activities.
  - Use non-traditional indicators (e.g., global economic policy uncertainty) as early warnings of system-wide financial risk build-up.
  - Elevated system connectedness should alert policymakers and regulators to heightened potential for build-up of risks in the global financial system.

*Source: Excerpt from "1. Sample of Global Systemically Important Banks and Insurers" (wp17210).*

### 1. Sample of Global Systemically Important Banks and Insurers .............................................8

### 1. Sample of Global Systemically Important Banks and Insurers

### Introduction
- Motivation: The Global Financial Crisis of 2007–2009 highlighted the role of global systemically important banks and insurers in the transmission of financial shocks across countries and markets.
- Regulatory context: Since 2011, the Financial Stability Board (FSB) has maintained lists of global systemically important banks (GSIBs) and global systemically important insurers (GSIIs).
- Focus: The paper examines interconnectedness among GSIBs and GSIIs as identified by the FSB in the past three years (published by the FSB in 2016, 2015 and 2014).

### Scope, data, and methodology
- Sample span: October 2007 to August 2016.
- Entities: U.S., European, and Asian GSIBs and GSIIs (as identified by the FSB lists referenced above).
- Market data used:
  - Publicly-available daily equity return data.
  - Intra-day equity volatility data.
- Primary empirical approach: Diebold and Yilmaz (2014 and 2015) framework applied to extract measures of connectedness (variance-decomposition based, mapping VAR variance decompositions to weighted directed networks).
- Comparative and complementary measures and methods referenced:
  - SRISK (Acharya et al. (2010, 2012)) — a balance-sheet based systemic risk measure.
  - Economic policy uncertainty (Baker, Bloom, and Davis (2016)).
  - Alternative interconnectedness measures in the literature mentioned: CoVaR (Adrian and Brunnermeier (2016)) and SyRIn (Segoviano et al. (2017, forthcoming)).
- Additional empirical tools:
  - Linear and quantile regression frameworks to analyze drivers of connectedness.
  - Scenario analysis to project connectedness conditional on factors such as global economic policy uncertainty and bond yields.
  - Panel data analysis to examine the role of bank-specific balance sheet factors (asset quality and profitability) in driving return connectedness for GSIBs.

### Main contributions of the paper
- Four stated dimensions of contribution:
  - Estimate market-based interlinkages among GSIBs and GSIIs to better understand connections among financial intermediaries, including the insurance sector’s role in propagating systemic financial risks.
  - Examine the relationship between the market-based connectedness measure (Diebold and Yilmaz) and the balance sheet-oriented SRISK measure.
  - Investigate the extent to which market-based connectedness can be explained by risk indicators such as economic policy uncertainty, VIX, and short and long maturity bond yields.
  - Analyze the role of bank balance sheet variables (profitability and asset quality) in driving bank-specific return connectedness.

### Key findings (as stated)
- Regional clustering:
  - Evidence of strong regional clusters in returns and volatility connectedness among GSIBs and GSIIs.
  - Strongest pairwise linkages are found within regions: Asia, the U.S., and Europe.
- Directionality:
  - Directional connectedness for global banks and insurers is from the United States and select European countries, to Asia.
  - Specific sources of outward returns and volatility connectedness: GSIBs from the United States, Germany, Spain, and France; GSIIs from the United States, France, and Germany.
  - Recipients of connectedness: GSIBs from China and Japan tend to be on the receiving end, based on market data.
- System-level dynamics:
  - The degree of total system connectedness—i.e., among all GSIBs and GSIIs in the sample—tends to rise during financial stress, coinciding with predictions of the balance-sheet oriented SRISK measure.
  - The evolution of the market-based interconnectedness measure is broadly in line with the SRISK measure.
- Role of macro-financial drivers:
  - Long term bond yields and global economic policy uncertainty play an important role in explaining system connectedness.
  - Rationale: Long term yields contain forward-looking information on economic growth; together with policy uncertainty, yield curve information influences pricing and trading decisions that affect connectedness.
- Bank-specific drivers:
  - Balance sheet variables such as asset quality and bank profitability play an important role in explaining market return connectedness for GSIBs.
- Caveat:
  - Findings should not be interpreted as general conclusions for the overall financial systems of countries where these institutions reside (e.g., GSIBs may represent a small subset of a country’s financial system).

### Analysis and projected workstreams referenced
- Comparison and validation:
  - The Diebold and Yilmaz connectedness measures will be compared with SRISK and considered alongside alternative methodologies in future research.
- Scenario analysis:
  - Scenario projections of connectedness conditional on macro factors (e.g., a U.S. growth scenario) are conducted later in the paper.
- Panel regressions:
  - Examination of bank-specific balance sheet factors in driving return connectedness for GSIBs via panel regression methods.

*Source: Excerpt from "1. Sample of Global Systemically Important Banks and Insurers" (wp17210).*

### 2014. The list covers United States, European, and Asian financial institutions that are

### wp17210 - 2014. The list covers United States, European, and Asian financial institutions that are

### Sample and data
- Sample period: daily equity returns and daily volatilities in equity prices from October 2007 to August 2016.
- Entities: 29 GSIBs and 10 GSIIs (39 financial institutions in total).
- Data sources: publicly-available equity price series from Datastream; assets data source notes: SNL and Moody’s where referenced.
- Treatment of missing data: remove a day if more than half of the entities have missing data; interpolate remaining missing observations.
- Time-zone adjustment: average two-day log returns for equity prices in local currency.

### Stylized facts: size, returns, and volatilities
- By assets, the largest GSIBs are located in Asia, followed by Europe and the United States.
- Asia example: Industrial and Commercial Bank of China (ICBC) and China Construction Bank combined asset value of US$6.5 trillion.
- Largest European banks named: HSBC and BNP Paribas.
- Largest U.S. banks named: JP Morgan Chase and Bank of America.
- Insurer vs bank assets:
  - While insurer asset size is lower on average, some GSIIs rival GSIBs.
  - AXA, Allianz and Metlife are comparable in size to Goldman Sachs, Morgan Stanley, and Credit Suisse.
- Average daily returns (sample average two-day log returns):
  - On average, European banks, Asian insurers, and U.S. insurers have the lowest (most negative) daily returns over the sample period.
  - General observation: daily equity returns for the global banks and insurers have been negative for the sample period from October 2007 to August 2016 despite episodes of recovery.
- Intra-day equity volatility (Parkinson and Diebold–Yilmaz methods; estimator and annualization):
  - Daily variance estimator: ߪ෤
௜௧
ଶ = 0.361 [ln(ܲ
௜௧
௠௔௫
) − ln(ܲ
௜௧
௠௜௡
)]^2
  - Annualized daily percent standard deviation (volatility): ߪො
௜௧
 = 100 √(365⋅ߪ෤
௜௧
 шал
).
  - Empirical averages: European banks and U.S. insurers have the highest intra-day volatility on average, followed by U.S. banks; Asian banks tend to have lower intra-day volatility.
  - Volatilities tend to be right-skewed; log volatilities are used to approximate normality.

### Methodology: measuring connectedness
- Core framework: Diebold and Yilmaz (2014, 2015) spillover analysis based on VAR and generalized forecast error variance decomposition (GVD).
- Advantages of the approach:
  - Unified, flexible framework to measure connectedness from pairwise to system-wide levels.
  - Measures can be static (full sample) or dynamic (time-varying).
  - Quantifies each institution’s contribution to network connectedness and systemic risk.
  - Enables estimation of cross-jurisdiction connections without entity-level cross-border exposure data.
  - Brings together properties of Marginal Expected Shortfalls and Delta CoVAR analogues.
- VAR and moving-average representation described with standard notation and assumptions of covariance stationarity.
- Use of generalized variance decomposition (Koop, Pesaran and Potter; Pesaran and Shin):
  - GVD is invariant to variable ordering and allows correlated shocks; suitable for large-N setting (39 institutions).
  - Pairwise directional connectedness ߠ̃
௜௝
௚
(ܪ) measures how much FI j explains FI i’s H-step-ahead variance.
- Aggregate connectedness measures:
  - From-degree (total directional connectedness of the system to FI i): ܥ
௜←	⋅
ு (expressed as percentage).
  - To-degree (total directional connectedness of FI i to the system): ܥ
⋅	←௜
ு (expressed as percentage).
  - Net-degree: difference between to-degree and from-degree; captures net contribution to systemic risk.
  - Total connectedness (system-wide): ratio of sum of off-diagonal elements of variance decomposition matrix to sum of all elements.
- Estimation for high-dimensional VAR:
  - Elastic net estimator (Zou and Hastie, 2005) used to blend shrinkage and selection, addressing the “curse of dimensionality” by combining lasso and ridge penalties.

### Empirical findings: dynamic connectedness and crisis episodes
- Total return and volatility connectedness tend to rise during financial crises and recessions.
- Global financial crisis (2007–2009):
  - Peak total connectedness in October 2008:
    - Returns: 92.5 percent
    - Volatility: 87.5 percent
  - Peak timing: shortly after Lehman Brothers filed for Chapter 11 bankruptcy protection (September 15, 2008).
- Greek sovereign debt crisis episode:
  - Connectedness rose to:
    - Returns: 89 percent (September 2010)
    - Volatility: 79 percent (September 2010)
  - Stayed high until February 2011.
  - As the bailout program for Greece restored market confidence, connectedness declined to:
    - Returns: 84.4 percent (summer 2011)
    - Volatility: 70.8 percent (summer 2011)
- Renewed European tensions (Italy, Spain, Portugal sovereign and banking issues):
  - Connectedness rose to:
    - Returns: 93.2 percent (beginning of [sample narrative context])
    - Volatility: 83.2 percent (beginning of [sample narrative context])

### Interpretations and implications
- The Diebold–Yilmaz connectedness metrics link to established systemic risk concepts:
  - From-degree analogous to Marginal Expected Shortfalls (exposure to systemic shocks).
  - To-degree analogous to Delta CoVaR (contribution to systemic events).
  - Net-degree identifies relative systemic importance (net transmitter vs net receiver).
- Dynamic upswings in connectedness correspond to acute stress events, reflecting heightened cross-institutional co-movement and potential amplification of shocks through equity markets.
- Use of equity-based measures and log-volatility transformations provides a data-driven, cross-jurisdictional lens on spillovers when balance-sheet level cross-border exposures are confidential.

*Italic: Source: wp17210 - 2014. The list covers United States, European, and Asian financial institutions that are (IMF working paper content provided).*

### 2012. European institutions undertook a number of policy measures to support and to clean

### wp17210 - 2012. European institutions undertook a number of policy measures to support and to clean

### Major policy measures and crisis episodes
- European long-term refinancing operation (LTRO) and LTRO2 provided European banks with long-term (three-year) funding at a low interest rate.
- The European Stability Mechanism (ESM) granted Spain a financial support package of €100 billion (10 percent of GDP) in June 2012.
- Return and volatility connectedness across the 39 financial institutions in Europe, the United States, and Asia declined to 88.4 percent and 70 percent, respectively, in October 2012.
- Volatility connectedness rose around the “fiscal cliff” (January 2013) and the debt ceiling debate (October 2013), peaking at a level of around 75 percent at the beginning of 2013.
- From July 2014, connectedness for both equity returns and volatilities began to increase gradually, reaching return and volatility connectedness of 87.6 percent and 77 percent, respectively, at the beginning of 2015.
- Notable episodes:
  - Global Financial Crisis (2007–2009): large spikes in connectedness.
  - European Sovereign Debt Crisis (peaked end-2013): elevated volatility connectedness.
  - Summer 2015: Chinese stock market disturbances with sharp declines in prices and rise in volatility.
  - Beginning of 2016: steep selloff in European bank stocks; connectedness among global banks and insurers rose sharply since January 2016 to levels observed during the European Sovereign Debt Crisis.

### Measurement approach and key estimation parameters
- Rolling total connectedness: rolling estimation window width is 200 days; predictive horizon for the underlying variance decomposition is 12 days.
- Rolling net directional connectedness: rolling estimation window width is 200 days; predictive horizon for the underlying variance decomposition is 12 days.
- Full sample period cited: October 2007 to August 2016.
- Sample of institutions: 39 financial institutions across Europe, the United States, and Asia (GSIBs and GSIIs).

### Pairwise and regional connectedness among GSIBs and GSIIs
- Regional clustering: strongest pairwise connectedness (top 15 percent) found within Asian, European, and U.S. clusters.
- China:
  - Chinese GSIBs: each explains around 19 percent of equity returns of the other institutions.
  - Chinese GSIIs’ equity returns could explain about 10 percent of equity returns of the Chinese GSIBs and vice versa.
- United States:
  - Two large U.S. GSIIs: each explains about 7 percent of equity returns of the other.
  - Among U.S. GSIBs: strongest connectedness among the four largest commercial banks at 6.2 percent on average; between the two large investment banks at 6.4 percent.
  - Highest bank-insurer connectedness among U.S. GSIB–GSII links at 5.3 percent.
- Europe:
  - Two Spanish banks: explain about 6.4 percent of each other’s equity returns.
  - Large French GSIBs: connectedness at 4.9 percent.
  - Two British insurers (European GSIIs): connectedness at 4.9 percent.
- Cross-region GSIB connectedness (for example U.S.–Europe) is generally less intense than within-region GSIB–GSII connectedness.
- Drivers of regional linkages include cross exposures, common investment exposure, client base, macroeconomic conditions, and financial regulations; insurers’ links to banks via investments, capital raising, and debt issuance.

### Net directional connectedness (system roles)
- By system-wide averages (October 2007 to August 2016):
  - U.S. banks and insurers appear to be the most important source of net directional connectedness to the system of GSIBs and GSIIs for both equity returns and volatilities.
  - Asian GSIBs and GSIIs tend to be recipients of net directional connectedness.
  - European GSIBs and GSIIs show more moderate average net directional connectedness but with significant cross-country variation.
- Dynamics (rolling sample):
  - U.S. banks and insurers were the dominant net contributors to return connectedness for the majority of the estimation period, with particularly elevated contributions during the 2007–2009 global financial crisis.
  - Dynamic volatility connectedness shows a more prominent role for European systemically important financial institutions relative to returns:
    - European GSIBs and GSIIs overtook U.S. counterparts in driving volatility connectedness in two main episodes:
      - Mid-2009 to end-2013 (European sovereign debt crisis).
      - Beginning of 2016 (sharp losses in European banking stocks).
    - By end of sample (August 2016), European banks and insurers were key drivers of volatility connectedness, with U.S. and Asian institutions as net receivers.

### Country-level and sub-sample findings (GSIBs and GSIIs)
- GSIBs (29 sub-sample):
  - Pairwise top 15 percent linkages concentrated among banks from the same region or country.
  - Strongest country groups: China, Japan, United States.
  - In Europe: strong return connectedness in Spain, France, United Kingdom, Switzerland.
  - Spanish, Italian, and French GSIBs explain more than 5 percent of each other’s equity returns on average.
  - Global banks from the United States and several European countries are main sources of return and volatility connectedness among GSIBs; Asian GSIBs are typically on the receiving end despite large asset size.
- Dynamic country-group patterns:
  - Net return connectedness elevated for GSIBs in Germany and the United States during the global financial crisis and European sovereign debt crisis.
  - Spanish and U.K. GSIBs were net contributors of return connectedness during these crises to a lesser extent.
  - Chinese and Japanese GSIBs are net recipients of system returns connectedness throughout the sample.
  - Dynamic volatility connectedness:
    - German and Spanish GSIBs exhibited elevated directional connectedness during the two crises.
    - U.S. GSIBs were net recipients of volatility connectedness during the European sovereign debt crisis.
    - Japan and China GSIBs were recipients of volatility connectedness throughout the sample.
- GSIIs:
  - European GSIIs highly interconnected among themselves; close linkages also between U.S. and European GSIIs.
  - Strong pairwise connectedness among two U.S. GSIIs; in Europe between French and Italian GSIIs and between German and U.K. GSIIs.
  - North American GSIIs are major contributors to both return and volatility connectedness; some European countries’ GSIIs also important contributors to return connectedness.
  - U.S. GSIIs were the most important net contributors of equity returns and volatility connectedness during the global financial crisis; spikes correspond to events such as the failure of Lehman Brothers (September 15, 2008) and the subsequent $180 billion bailout of AIG by the Federal government.
  - During the European Sovereign Debt Crisis, German and French GSIIs were main sources of return and volatility connectedness.

### Robustness checks and sensitivity
- Robustness strategy:
  - Condition institution-specific returns on global and country-specific returns and bond yields, using residuals of filtered series.
  - Use alternative market prices: Credit Default Swaps (CDS) spreads (senior five-year spreads from Datastream), in log-difference.
  - Vary empirical estimation assumptions: estimation window of 150 days in addition to 200 days; lag order from two to three lags; alternative forecast horizons for variance decomposition.
- CDS coverage note: CDS coverage for banks and insurers is lower than equity prices; estimation sample for CDS spans from December 2007 to August 2016. Institutions excluded from CDS estimation due to lack of CDS data or short coverage: Mitsubishi, ICBC, China Construction Bank, Bank of New York Mellon, State Street, UniCredit, Nordea, Credit Suisse and Ping An Insurance.
- Main robustness conclusion: total connectedness index remains largely unchanged across robustness checks, with elevated levels of system interconnectedness apparent during the global financial crisis, the European sovereign debt crisis, and in the recent period examined.

### Comparison with SRISK (balance sheet-oriented systemic risk measure)
- SRISK overview:
  - Measures expected capital shortfall if firms’ equity values were to decline to GFC levels; aggregate across largest financial firms (banks and nonbanks, generally insurers).
  - SRISK computation elements:
    - Long run marginal expected shortfall (LRMES) uses a six-month crisis threshold for the broad market index decline, taken to be 40 percent.
    - ߚ is the firm’s CAPM-based beta coefficient.
    - Prudential capital requirement k set at 8 percent for firms in Africa, Asia, and Americas, and 5.5 percent for firms in Europe.
    - Approximate SRISK formula involves k, current EQUITY (market capitalization), and DEBT (book value of assets minus book value of equity).
- Empirical comparison:
  - Global connectedness (market-information-based) plotted against SRISK for a global system including 11 domiciliary countries, on a bi-annual frequency.
  - Broad trends in the connectedness measure accord closely with SRISK.
  - SRISK is increasing in the correlation between individual firms’ equity returns and the broad market index (captured in beta ߚ); capital requirement k affects overall level of SRISK.

### Common factors and drivers of global connectedness
- Market risk indicators considered: MOVE index (fixed-income implied volatility), VIX index (S&P 500 implied volatility), U.S. Dollar Index (DXY), and global economic policy uncertainty (GPU) index (GDP-weighted PPP-adjusted average for 16 countries).
- Observed relationships (visual inspection and correlations):
  - Co-movement between global connectedness and MOVE and VIX is not consistently strong, particularly since mid-2014.
  - Global connectedness moved consistently in line with global economic policy uncertainty (GPU), showing pro-cyclicality on average over the sample.
  - Upswing in connectedness since late 2014 accompanied by steady appreciation of the U.S. dollar (DXY).
- Bond yields:
  - Comparison of global connectedness with 3-month and 10-year bond yields for illustrative countries (China, Germany, United States):
    - 10-year yields display a countercyclical relationship with connectedness; Germany shows a more persistent downward trend in 10-year yields.
    - Relationship between 3-month yields and connectedness is heterogeneous across countries.
  - Principal components analysis of 10-year yields across 11 countries:
    - First principal component (PC) moves in line with the U.S. 10-year yield (US10), with more pronounced cyclical amplitude.
    - Subsequent analysis focuses on US10 as a common factor proxy.
- Formal tests:
  - Pair-wise correlations: GPU most strongly correlated with global connectedness in absolute terms; DXY the weakest.
  - Dynamic impulse response analysis: global connectedness responds positively to a shock to GPU and negatively to a shock to US10 across the entire sample and two subsamples.

*Source: IMF staff estimates and analysis as presented in the provided content.*

### Box 1. Dynamic Responses of Global Connectedness to Selected Shocks

### Box 1. Dynamic Responses of Global Connectedness to Selected Shocks

### VAR impulse-response analysis: setup and identification
- Three bivariate VAR systems estimated: (i) GPU and GC; (ii) US10 and GC.  
- Direction of contemporaneous causality assumed from GPU and US10 towards connectedness; VARs identified via a Cholesky scheme.  
- VIX included as an exogenous variable to control for broad based market risk.  
- Three estimation samples:
  - Full sample: July 2008–August 2016
  - Subsample A: July 2008–August 2012
  - Subsample B: August 2012–August 2016
- Impulse responses trace the dynamic response of connectedness to a one standard deviation shock (increase) in (i) global economic policy uncertainty (GPU), and (ii) the U.S. 10 year (spot) yields (US10).
- Time horizon for responses: 24 months (horizontal axis). Units on the vertical axis: standard deviations from mean.
- 95 percent confidence bands computed via 20,000 Monte Carlo simulations.
- Main finding:
  - Connectedness has responded significantly to increases in GPU and US10 across the entire sample and both subsamples.
  - Directions of contemporaneous responses accord with Figures 18, 19 and Appendix Table 1 (as reported in source).
  - Impulse responses remain qualitatively similar if VIX is not controlled for, or if a ‘generalized’ identification scheme is used.

### Predictive regression (specification C1): model, estimation, and inference
- Predictive model (C1): GC_t = alpha + beta·GPU_t + sigma·US10_t + phi·VIX_t + theta·GC_{t-1} + epsilon_t, with epsilon_t allowed a GARCH(1,1) disturbance:
  - epsilon_t ~ N(0, s_t)
  - s_t = a0 + a1·epsilon_{t-1}^2 + b1·s_{t-1} (following Bollerslev (1986))
- Conditioning variables chosen based on strength of co-movement with GC. GPU, US10, and VIX assumed to affect GC contemporaneously. One-period lagged GC included to capture persistence. Lags for other variables were found insignificant and dropped.
- Rationale for VIX: control for volatility in equity markets and proxy for risk aversion (Bekeart and Hoerova, 2016).

### Linear regression estimation results (July 2008 – August 2016)
- All conditioning variables significant at the 5 percent level.
- Overall R-squared = 0.89.
- Pro-cyclicality of GPU and counter-cyclicality of US10 with respect to GC is evident.
- No evidence of time-varying heteroscedasticity governing the evolution of GC was found.
- Variance inflation factors for included variables were less than 2 in all cases (evidence against serious collinearity).

Key estimated coefficients and statistics (as reported in Table 2)
- Linear regression:
  - alpha = 21.65** (4.50)
  - beta (GPU) = 0.01** (0.00)
  - sigma (US10) = -0.52** (0.26)
  - phi (VIX) = 0.04** (0.02)
  - theta (lagged GC) = 0.74** (0.05)
  - a0 = 1.01 (0.80)
  - a1 = 0.58 (0.39)
  - b1 = 0.33 (0.61)
  - R-squared = 0.89
- Note: (**) denotes significance at the 5 percent level. Standard errors in parentheses.

### Quantile regression findings (based on specification C1)
- Quantile regression estimated across τ = 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9 (sparsity method based on Epanechnikov kernel; standard errors computed via bootstrap with 20,000 residual resamples).
- US10:
  - Although countercyclical with respect to GC in conditional mean terms, US10 has a more significant impact on higher percentiles of the conditional distribution of GC; i.e., τ ≥ 0.60.
- GPU:
  - Exhibits a more uniform impact—in magnitude and statistical significance—across the range of percentiles.
- Selected quantile coefficient examples (from Table 2):
  - alpha at τ = 0.1: 28.83** (7.35)
  - beta (GPU) across quantiles: 0.01 (0.00) with ** at multiple quantiles
  - sigma (US10) at τ = 0.9: -1.08** (0.18)
  - phi (VIX) at τ = 0.9: 0.09** (0.01)
  - theta (lagged GC) ranges, e.g., 0.64** (0.08) at τ = 0.1 and 0.76** (0.03) at τ = 0.9

### Within-sample and out-of-sample predictive performance
- Within-sample predictive accuracy:
  - One-step ahead predictions computed over estimation sample July 2008 to August 2016.
  - Standard error bands computed via bootstrap resampling of the error term; 5,000 bootstrap resamples used at each time-step for one-step ahead conditional mean predictions.
  - Model shows high degree of within-sample predictive accuracy when comparing one-step ahead predictions with actual global connectedness.
- Out-of-sample predictive accuracy:
  - Model re-estimated over July 2008 to April 2016 (prediction origin April 2016). Prediction horizon: April 2016 to August 2016.
  - Conditioning on observed exogenous variables over April 2016 to August 2016, dynamic out-of-sample predictions track broad trends in GC reasonably well.

### Counterfactual analyses (April 2016 origin; conditional paths August 2016–January 2017)
- Observed conditioning-path (August 2016 to January 2017):
  - US10 increased from 1.55 percent in August 2016 to 2.43 percent by January 2017.
  - GPU increased from an index level of 175.73 to 270.15 over the same period (reported as a 70 percent increase in text).
  - VIX exhibited fairly low variability, hovering close to its average value recorded over the preceding six-month period.
  - Predicted path for GC oscillated slightly around the level at the prediction origin (August 2016).
- Counterfactual 1:
  - Assume rise in US10 accompanied by flat-lining of GPU at its average value over 12 months prior to prediction origin (approximately equal to 166).
  - Conditioning on this path (and observed values for remaining variables) would set GC on a downward trajectory.
- Counterfactual 2:
  - Question: what magnitude of increase for US10 would be needed to counteract the effect of the observed 70 percent increase in GPU (from 175.73 to 270.15), so that predictions of GC align with the downward trajectory from Counterfactual 1?
  - Reported conclusion: to generate the same downward trajectory for GC, US10 would need to have risen extremely (and implausibly) steeply (text ends here in source).

*Source: Bloomberg, Bankscope, FRED, IMF Staff estimates*

### 1.55 percent to near 4 percent.

### 1.55 percent to near 4 percent.

### Out-of-Sample Predictions and Counterfactuals
- Estimation sample: July 2008 to August 2016, with August 2016 the out-of-sample prediction origin.
- Figure 23 presents out-of-sample predictions and two counterfactuals for Global Connectedness (GC), conditional on alternative paths for conditioning variables (see Box 2 for long-term projection scenario).
- Key methodological point: global connectedness is the aggregate of to- and from-degrees. The to-degree (ܥܩ_ܶ
௝,௧) denotes the total directional connectedness of GSIBs (GSIIs) in a particular region “to” the system; the from-degree (ܥܩ_ܨ
௝,௧) denotes total directional connectedness “from” the system to that region.

### Predictive Regression Results — Disaggregated (Table 3) and Heterogeneity
- Sample span for estimation: July 2008–August 2016. Linear regressions estimated via maximum likelihood. Standard errors in parentheses. (**) and (*) indicate significance at the 5 percent and 10 percent significance levels, respectively.
- Aggregate versus disaggregated dynamics:
  - For the aggregate GC measure, GARCH effects were relatively unimportant (see referenced Table 2).
  - At the disaggregated level (to- and from-degrees), GARCH (1,1) effects are very relevant, indicating significant idiosyncratic firm- or region-specific dynamics.
- Representative coefficients from Table 3 (predictive regression for total directional connectedness — To- and From-degrees). Coefficients marked ** and * indicate significance at the 5 percent and 10 percent levels, respectively.
  - To-degree (ܥܩ_ܶ
௝,௧), GSIBs:
    - U.S. alpha (ߙ): 46.90** (6.57)
    - Europe alpha (ߙ): 18.81** (5.06)
    - Asia alpha (ߙ): 3.46 (3.30)
    - Beta (ߚ) examples: U.S. -0.01 (0.00); Europe 0.01 (0.01); Asia 0.02 (0.00)
    - Sigma (ߛ) examples: U.S. -3.95** (0.93); Europe -0.56 (0.60); Asia 1.78** (0.71)
    - Phi (ߜ): U.S. 0.29** (0.05); Europe -0.04** (0.02); Asia 0.04 (0.20)
    - Theta (ߠ): U.S. 0.58** (0.04); Europe 0.82** (0.05); Asia 0.73** (0.06)
    - GARCH parameters (߱଴, ߱ଵ, ߱ଶ) examples: ߱଴ U.S. 10.68** (3.37), Europe 1.55** (0.12), Asia 4.60** (1.99); ߱ଶ reported as 0.69** (0.25) etc.
  - To-degree (ܥܩ_ܶ
௝,௧), GSIIs:
    - U.S. alpha (ߙ): 47.87** (5.94)
    - Europe alpha (ߙ): 9.72** (4.70)
    - Asia alpha (ߙ): 1.92 (4.41)
    - Beta (ߚ) examples: Europe 0.04** (0.00)
    - Sigma (ߛ) examples: U.S. -4.37** (0.75); Europe 1.63** (1.28)
    - Phi (ߜ): U.S. 0.24** (0.08); Europe 0.09** (0.04); Asia 0.06 (0.01)
    - Theta (ߠ): U.S. 0.61** (0.06); Europe 0.82** (0.06); Asia 0.80** (0.02)
    - GARCH parameters (examples): ߱଴ U.S. 30.07** (8.57), Europe 0.74** (0.56), Asia 4.79 (5.41); ߱ଶ entries include 0.78** (0.33), 0.19 (0.15), 0.71** (0.28) etc.
  - From-degree (ܥܩ_ܨ
௝,௧), GSIBs:
    - U.S. alpha (ߙ): 27.60** (4.30)
    - Europe alpha (ߙ): 21.77** (3.29)
    - Asia alpha (ߙ): 28.19** (4.12)
    - Beta (ߚ) examples: U.S. 0.01** (0.00); Europe 0.01** (0.01)
    - Sigma (ߛ): U.S. -0.98** (0.17); Europe -0.84** (0.17); Asia -1.54** (0.48)
    - Phi (ߜ): U.S. 0.03** (0.00); Europe 0.07** (0.01); Asia 0.09** (0.03)
    - Theta (ߠ): U.S. 0.71** (0.04); Europe 0.76** (0.04); Asia 0.66** (0.05)
    - GARCH parameters (examples): ߱଴ U.S. 0.29** (0.09), Europe 0.47** (0.17), Asia 2.50** (0.83); ߱ଶ entries include 0.83** (0.36), 0.06 (0.03), 0.81** (0.23) etc.
  - From-degree (ܥܩ_ܨ
௝,௧), GSIIs:
    - U.S. alpha (ߙ): 22.37** (7.20)
    - Europe alpha (ߙ): 15.56** (4.04)
    - Asia alpha (ߙ): 17.61** (6.07)
    - Beta (ߚ): Europe 0.02** (0.00)
    - Sigma (ߛ): U.S. -1.24** (0.28); Europe 0.17** (0.19); Asia 0.89 (1.19)
    - Phi (ߜ): Asia 0.09 (0.06)
    - Theta (ߠ): U.S. 0.78** (0.08); Europe 0.78** (0.05); Asia 0.71** (0.07)
    - GARCH parameters (examples): ߱଴ U.S. 0.13 (0.16), Europe 0.16** (0.00), Asia 1.54** (0.21); ߱ଶ entries include 0.15** (0.12), 0.91** (0.33), 0.00 (0.11) etc.
- Interpretation:
  - There is heterogeneity across regions and firm types in both direction and significance of predictors.
  - Idiosyncratic, firm- or region-specific factors appear to play a significant role in to- and from-degree dynamics alongside systemic factors such as global economic policy uncertainty (GPU) and U.S. 10-year yield (US10).
  - At aggregate level, dynamics of GC can be explained almost exclusively by systemic factors—U.S. long-term bond yield and global economic policy uncertainty.

### The Role of Bank-Specific Factors — Panel Regression (Table 4)
- Estimator and data:
  - Panel regression for to- and from-degree connectedness for GSIBs estimated with the Arellano-Bover/Blundell-Bond linear dynamic panel estimator with robust standard errors.
  - Sample span: 2008 to 2015. Data sources include FitchConnect and policyuncertainty.com.
  - Number of Bank_id: 29. Observations: 195 for each regression.
- Specification includes lagged dependent variable, bank-fixed effects, and time-varying bank-specific indicators: ROAA (profitability), NPL ratio (asset quality), and total assets (ln). International/global indicators include global economic policy uncertainty and U.S. 10-year yield.
- Selected coefficient estimates (standard errors in parentheses; significance: ** p<0.05, *** p<0.01, * p<0.1):
  - To_Connectedness (column 1):
    - Lagged dependent variable: 0.467*** (0.105)
    - ROAA: -7.905** (3.466)
    - NPL ratio: 1.077 (1.189)
    - Total assets (ln): -3.148 (5.392)
    - Global uncertainty: 0.0624 (0.0761)
    - US 10-year interest rate: 0.0173 (2.540)
    - Constant: 80.37 (81.62)
  - From_Connectedness (column 2):
    - Lagged dependent variable: 0.229** (0.0928)
    - ROAA: -4.807*** (1.770)
    - NPL ratio: 0.906* (0.485)
    - Total assets (ln): -1.666 (1.033)
    - Global uncertainty: 0.130*** (0.0267)
    - US 10-year interest rate: 1.948* (1.084)
    - Constant: 65.51*** (18.08)
- Key findings from the panel regression:
  - Bank profitability (ROAA) plays an important role: less profitable GSIBs are both sources of return connectedness (outward) and more prone to contagion (inward).
  - Higher NPL ratio (poorer asset quality) is associated with higher inward connectedness (contagion).
  - Elevated global uncertainty and higher US10-year interest rate increase inward connectedness (contagion risk) on average.

### Conclusions and Policy Implications
- Data and methods:
  - Connectedness measured from forecast error variance decomposition of a VAR model using the Diebold and Yilmaz approach, based on publicly available equity returns and intra-day volatility from October 2007 to August 2016.
- Empirical conclusions:
  - GSIBs in the United States and several European countries are important sources of directional equity return and volatility connectedness.
  - GSIIs from the United States and some European countries also tend to influence other GSIIs’ equity returns and volatilities.
  - Largest GSIBs appear to be recipients of connectedness, while largest GSIIs tend to be sources of outward returns and volatility connectedness.
  - Strong regional clusters of return and volatility connectedness exist among Asian, United States, and European GSIBs and GSIIs; cross-sectoral connectedness within the same region is more intense than cross-regional linkage within sectors.
  - Total system connectedness tends to rise during financial stress episodes (e.g., Global Financial Crisis, European Sovereign Debt Crisis).
  - Higher global economic policy uncertainty and lower U.S. 10-year bond yields are associated with increased market connectedness among GSIBs and GSIIs.
  - Bank-specific factors (profitability, asset quality) significantly affect return connectedness.
- Policy implications:
  - The prominence of GSIBs from the United States, Germany, Spain, and France highlights the need for strong risk management and intense supervision of global systemically-important institutions, including monitoring cross-border exposures.
  - GSIIs from France, the United States, and Germany similarly require intensive supervision with attention to inward and outward spillovers.
  - High cross-sectoral connectedness between banks and insurers emphasizes monitoring of cross-sectoral exposures and comprehensive group-wide supervision that accounts for all risk activities across a financial conglomerate and its subsidiaries.
  - Increases in system connectedness should alert policymakers and regulators to heightened potential for build-up of risks in the global financial system.
  - ‘Non-traditional’ indicators, such as global economic policy uncertainty, can convey useful information on system-wide financial risks.

### Box 2 — Projecting Global Connectedness Conditional on a U.S. Growth Scenario
- Conditioning approach:
  - Extended paths for conditioning variables calibrated from historical behavior during September 2002 to December 2004. This period was selected because U.S. growth increased over subsequent years to be close to 4 percent after starting from a point close to the 2016 level.
  - Over September 2002 to December 2004, the U.S. 10-year yield maintained a general upward trend from 3.87 percent to 4.23 percent.
  - At the prediction origin (January 2017), the level of yields is considerably lower than at September 2002. The conditioning path for US10 is constructed by adapting the trend based on the average growth rate in yields over the historical period to circumvent a level shift.
- Global economic policy uncertainty (GPU) conditioning:
  - Three GPU conditioning paths considered: historical maximum, minimum, and median levels observed over the 12 months preceding January 2017.
  - VIX level taken to be the 2002–2004 average.
  - Dashed lines in the figure trace predictions for GC conditional on the median GPU measure, with upper and lower bounds defined by the max/min GPU conditioning paths.
- Main projection takeaway:
  - Conditioning on the US10 trajectory informed by the U.S. growth scenario and the median GPU path, a fall in global connectedness to its early 2015 level can be expected over a horizon of 2.5 years, approximately.

*Source: Bloomberg, Bankscope, FRED, FitchConnect, http://www.policyuncertainty.com/, IMF Staff estimates*

### Chapter 1.

### Chapter 1.

### References
- GFSR (April 2016a), “The Growing Importance of Financial Spillovers from Emerging Market Economies” GFSR Report, Chapter 2.
- GFSR (April 2016b), “The Insurance Sector—Trends and Systemic Risk Implications” GFSR Report, Chapter 3.
- Guimaraes-Filho, Roberto and Hong, Gee Hee (2016), “Dynamic Connectedness of Asian Equity Markets,” IMF Working Paper 16/57.
- International Association of Insurance Supervisors (2011), “Insurance and Financial Stability”, November 2011.
- Jackson, M.O. (2008), “Social and Economic Networks,” Princeton University Press.
- Newman, M.E.J. (2010), Networks: An Introduction, Oxford University Press.
- Ohnsorge, F. L., M. Wolski, and Zhang, Y. S. (2014), “Safe Havens, Feedback Loops, and Shock Propagation in Global Asset Prices,” IMF Working Papers 14/81, International Monetary Fund
- Pearl, J. (2000), Causality: Models, Reasoning, and Inference, Cambridge University Press.
- Segoviano, M., S. Malik, P. Lindner, and P. Cortes (2017), “A Comprehensive Multi-Sector Framework for Surveillance of Systemic Risk and Interconnectedness (SyRIN)”, IMF Working Paper, forthcoming.
- Sims, C. A. (2002), “Solving Linear Rational Expectations Models,” Computational Economics, 20(1-2),1–20.
- Stock, J. H., and Watson, M.W. (1989), “New Indexes of Coincident and Leading Indicators.” In: Blanchard, O., Fischer, S. (Eds.), NBER Macroeconomics Annual. MIT Press, Cambridge, MA.
- Pesaran, M. Hashem, and Shin, Yongcheol (1996), “Cointegration and Speed of Convergence to Equilibrium," Journal of Econometrics, 71(1-2), 117-143.
- Porter, Nathan, and Xu, TengTeng (2006), "Money-Market Rates and Retail Interest Regulation in China: The Disconnect between Interbank and Retail Credit Conditions," International Journal of Central Banking, vol. 12(1), pages 143–198, March 2016.
- White, H. and K. Chalak (2009), “Settable Systems: An Extension of Pearl's Causal Model with Optimization, Equilibrium, and Learning," Journal of Machine Learning Research, 10, 1759–1799.

### Appendix figures (titles and captions preserved)
- Appendix Figure 1. Net Dynamic Return Spillover to the System of GSIBs
- Appendix Figure 2. Net Dynamic Volatility Spillover to the System of GSIBs
- Appendix Figure 3. Net Dynamic Return Spillover to the System of GSIIs
- Appendix Figure 4. Net Dynamic Volatility Spillover to the System of GSIIs
- Appendix Figure 5. Robustness Check—Alternative Market Prices
  - Total Connectedness: CDS Spreads
  - Total Connectedness: Equity Volatility
- Appendix Figure 6. Robustness Check—Estimation Window
  - Total Return Connectedness: Estimation window =150 days
  - Total Return Connectedness: Estimation window =200 days
- Appendix Figure 7. Robustness Check—Lag Order
  - Total Volatility Connectedness: Lag Order =2
  - Total Volatility Connectedness: Lag Order =3
- Appendix Figure 8. Robustness Check—Forecast Horizon
  - Total Return Connectedness: Forecast Horizon = 12 Days
  - Total Return Connectedness: Forecast Horizon = 24 Days
- Appendix Figure 9. Global Connectedness vs. Three-Month Yields
  - Source: Bloomberg, Bankscope, FRED, IMF Staff estimates. Note: ‘rhs’ refers to right hand side axis.
  - Panels: Global ConnectednessChina (rhs); Global ConnectednessNetherlands (rhs); Global ConnectednessFrance (rhs); Global ConnectednessSpain (rhs); Global ConnectednessGermany (rhs); Global ConnectednessSweden (rhs); Global ConnectednessJapan (rhs); Global ConnectednessSwitzerland (rhs); Global ConnectednessItaly (rhs); Global ConnectednessU.K. (rhs); Global ConnectednessU.S. (rhs)
- Appendix Figure 10. Global Connectedness vs. Ten-Year Yields
  - Source: Bloomberg, Bankscope, FRED, IMF Staff estimates. Note: ‘rhs’ refers to right hand side axis.
  - Panels: Global ConnectednessChina (rhs); Global ConnectednessNetherlands (rhs); Global ConnectednessFrance (rhs); Global ConnectednessSpain (rhs); Global ConnectednessGermany (rhs); Global ConnectednessSweden (rhs); Global ConnectednessJapan (rhs); Global ConnectednessSwitzerland (rhs); Global ConnectednessItaly (rhs); Global ConnectednessU.K. (rhs); Global ConnectednessU.S. (rhs)

### Appendix Table 1. Global Connectedness Pair-Wise Correlations (Sample: July 2008–August 2016)
- Header: GPU SLOPE US10 VIX US3 MOVE DXY
- Row: 0.57 -0.40 -0.38 0.29 0.22 0.21 0.15
- Source: Bloomberg, Bankscope, FRED, IMF Staff estimates
- Note: Sample: July 2008–August 2016.

*Source: wp17210 - Chapter 1 (PDF).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17210.pdf_