Euro Area Policies: Financial Sector Assessment Program Technical Note—Systemic Risk Analysis

International Monetary Fund. Monetary and Capital Markets Department

doi:10.5089/9781484369586.002.A003

Author:

International Monetary Fund. Monetary and Capital Markets Department

International Monetary Fund. Monetary and Capital Markets Department
Search for other papers by International Monetary Fund. Monetary and Capital Markets Department in
Current site
Google Scholar

Publication Date:: 19 Jul 2018

eISBN:: 9781484369586

Language:: English

Keywords:: ISCR; CR; bank; OFI residual; other financial institution; NPL ratio; OE banking systems; net spillover; spillovers vis-à-vis banking systems; extra-EA banking systems; banking spillover

Download PDF (1.9 MB)

Financial Sector Assessment Program-Technical Note-Systemic Risk Analysis

Abstract

Financial Sector Assessment Program-Technical Note-Systemic Risk Analysis

Market-Based Interconnectedness Analysis³²

A. Introduction

63. Banking system interconnectedness is central to financial stability. It features prominently in key aspects of systemic risk, that is, the system-wide connectedness. Indeed, interconnectedness, along with size and complexity, for example, is one of the main criteria used by the Financial Stability Board (FSB) when determining globally systemically important banks (G-SIBs).

64. In this chapter, market-based spillover measures are used to gauge the interconnectedness of the euro area banking system. Equity price data are used to estimate spillovers between euro area and non-euro area listed banks across 28 countries over 2006–2017. While bank-level data are used, the focus is on net spillovers (the difference between outward and inward spillovers) from the perspective of the euro area banking system vis-à-vis selected non-EA banking systems. This focus is partly motivated by the fact that the EA has one of the largest, most complex, and globally interconnected banking systems in the world; and is home to several G-SIBs.

65. Our proposed methodological framework is novel, in that it entails deriving conditional net spillover distributions. Specifically, quantile regressions are used to generate distributions of net spillovers conditional on banking system characteristics (for example, solvency, profitability, and asset quality metrics) and macroeconomic conditions (for instance, real GDP growth). This allows us to gauge how changes in such determinants influence the entire distribution of net spillovers. Put differently, going beyond the mean, the chapter quantifies how shifts in selected bank characteristics affect the variance, skewness, kurtosis, and hence more generally, the conditional distributions of net spillovers. The framework thus stands in contrast to more traditional linear empirical strategies which, conditional on determinants, seek to analyze net spillovers solely in terms of the mean outcome.³³

66. While the a priori link between bank-specific characteristics and average spillovers maybe relatively intuitive, the degree to which such characteristics influence the conditional distribution of net spillovers is less clear, warranting an empirical investigation. For example, a natural conjecture would be that greater capitalization would reduce inwards spillovers on average. However, the implications for higher moments, and the distribution of (net) spillovers is less clear. Likewise, some determinants may primarily shift the conditional distribution’s central tendency (including the average), while others may tend to operate more on the tails, that is, largely impacting tail risks. Classifying determinants based on how they impact different segments of the distribution could be informative from a risk management perspective. Importantly, the proposed framework allows formal quantification of the likelihood of a banking system being a recipient or transmitter of spillovers, and how this likelihood would vary as changes in various determinants (such as capital buffers, profitability, and asset quality metrics) shift the conditional distribution of net spillovers.

67. The main findings can be summarized as follows:

The EA banking system is equally as likely to be a recipient of inward spillovers from the U.S. and other European banking systems as it is to be a transmitter of outward spillovers to these systems. However, given its more leptokurtic shape of the distribution, EA net spillovers vis-à-vis extra-EA banking systems are more prone to tail risks. The EA appears much more likely to be a transmitter of outward spillovers to banking systems in emerging and (other) advanced economies, than being a recipient
Stronger bank fundamentals reduce the likelihood of inward spillovers from the rest of the world not only on average, but also in terms of tail risks. In particular, lower NPL ratios, greater profitability, and higher capitalization levels are shown to decrease the probability of inward spillovers to the euro area banking system from the rest of the world.
Evidence over the past five years suggests stronger bank fundamentals (such as better asset quality), may reduce inwards spillovers to a greater degree relative to earlier in the decade.
Evidence is found that progressively stronger fundamentals can enhance the euro area banking system’s resilience to inward spillovers without necessarily aggravating outward spillovers. Thus, strong euro area fundamentals appear to enhance the stability of other regions.

B. General Framework: An Overview

This section begins with a summary of the market-based spillovers measure used, and then moves on to the novel contribution of the chapter which is the derivation of probability distribution of net spillovers conditional on selected determinants.

Measuring Spillovers

68. Within the broader context of market-based interconnectedness, spillovers could be considered a directional concept. At any point in time, a financial entity can simultaneously act as a potential transmitter of shocks to market prices (outward spillovers) and receiver of shocks (inward spillovers), vis-à-vis other entities in an interconnected system. A “net” measure of spillovers for a particular entity, conceptualized as the difference in magnitudes between outward and inward spillovers, serves to capture the balance of risks associated with these market-based interconnectedness metrics. Given such a construct, negative net spillovers imply that the entity under consideration is a net receiver of spillovers, whereas a positive measure would correspond to the entity acting as a net transmitter.³⁴

69. From a practical standpoint, one particular way to obtain market-based spillovers measures is via the Diebold-Yilmaz (2014, 2015) approach (DY, henceforth). This approach relies on estimating a vector autoregressive (VAR) model, employing time series data for a set of banks (in this case, on daily equity returns). Within the VAR framework, spillover metrics are derived via a forecast error variance decomposition, which quantifies the proportion of return variability (contemporaneous and H-steps ahead) of a particular bank i, that can be attributed to shocks to returns of another bank j, for instance.³⁵ This quantity, $C_{i \leftarrow j}^{H}$ say, is taken to proxy the spillover from j to i. Conversely, the proportion of j’s return variability, given shocks to bank i’s returns can also be computed. This would correspond to spillover from i towards j, i.e., $C_{i \to j}^{H}$ . It follows that a net spillover measure for bank i vis-à-vis j would simply be the difference:

\begin{matrix} n e t s p i l l o v e r_{i \leftrightarrow j} = C_{i \to j}^{H} - C_{i \leftarrow j}^{H} . & (1) \end{matrix}

Generating Conditional Distributions

70. Assuming availability of net spillover measure, the link with determinants can be made using quantile regressions (Koenker and Bassett, 1978 and Koenker, 2005). For net spillover series corresponding to a single bank i vis-à-vis another bank j the equation to be estimated could be written follows:

\begin{matrix} (n e t s p i l l o v e r_{i \leftrightarrow j, t}^{q} =) y_{(i), t}^{q} = α + β^{q} B a n k_{(i), t} + γ^{q} M a c r o_{(i), t} + ε_{t}^{q} & (2) \end{matrix}

where the time subscript t (=1,...,T), denotes quarters; Bank denotes a B × T vector of bank i’s balance sheet characteristics; Macro denotes an M × T vector of macroeconomic conditions corresponding to the country where bank i is headquartered; and q denotes various percentiles of interest for which equation (2) is to be estimated, that is, q = {0.05; 0.25; 0.50; 0.75; 0.95}. The estimated conditional quantile function (inverse cumulative distribution function) would in turn correspond to: ${\hat{y}}_{(i), t}^{q} (= α + {\hat{β}}^{q} B a n k_{(i), t} + {\hat{γ}}^{q} M a c r o_{(i), t})$ .

71. Given the noisiness of quantile functions estimates in practice, recovering the corresponding PDF will require smoothing of the quantile function. In this chapter, in line with the approach of Adrian, Boyarchenko, and Giannone (2017), (see also GFSR 2017), this is accomplished via fitting a (parametric form) ‘skewed’ t-distribution:

\begin{matrix} f (y; μ, s, v, ξ) = {\begin{matrix} \frac{2}{ξ + \frac{1}{ξ}} ξ g (z), & z < 0 \\ \frac{2}{ξ + \frac{1}{ξ}} g (z) / ξ, & z \geq 0 \end{matrix} & (3) \end{matrix}

where $g (z) = \frac{1}{s} \bar{g} (z; ν)$ , with $\bar{g} (.)$ denoting the PDF of standard Student-t with ν degrees of freedom; z is given by $\frac{y - μ}{s}$ with μ and s referring to location and scale parameters, respectively. Skewness is governed by shape parameter ξ. This functional form for the skewed t-distribution is based on that motivated by Fernandez and Steel (1998), further explored and refined in Giot and Laurent (2003) and Lambert and Laurent (2002); see also Boudt, Peterson and Croux (2009).³⁶ For specified values for the conditioning variables (or point in time), four parameters {μ, s, ν, ξ} of the implied density determined by minimizing the squared distance between the estimated quantile function, ${\hat{y}}^{q}$ , and theoretical quantile function y^q,f (μ, s, ν, ξ) corresponding to the above skewed-t distribution (see Giot and Laurent, 2003). Specifically, the 5th, 25th, 50th, 75^th, and 95th percentiles can be matched via distance minimization:

\begin{matrix} {μ, s, ν, ξ} = \underset{μ, s, ν, ξ}{\arg \min} \sum_{q} {{\hat{y}}^{q} - y^{q, f} (μ, s, ν, ξ)}^{2} & (4) \end{matrix}

where μ ∈ ℝ, s > 0, ν ≥ 2 and ξ > 0. Notwithstanding the skewness property, the choice of a skewed-t functional form is advantageous from the perspective of flexibility. For example, as ν → ∞, f(y; μ, s, ν, ξ) is characterized by tail properties resembling a Gaussian; moreover, the density is symmetric for ξ = 1.

C. Data and Stylized Facts

This section reviews the publicly-available data used to estimate spillovers and some stylized facts, briefly presents an example illustrating how net spillovers have evolved over time, and then discusses the possible determinants of net spillovers.

Market-based Data

72. Market-based spillovers are derived using daily returns from equity price data for a sample of 93 global banks (Appendix II, Table 1). These banks are allocated to five regional banking systems: euro area (EA), Other Europe (OE), the United States (U.S.), Advanced Economies, excluding U.S. and Europe (AE), and emerging markets (EM). In particular, OE includes banks headquartered in Sweden and the United Kingdom, whereas AE includes Australian, Canadian, and Japanese banks (Appendix II, Table 1). The sample is restricted only to banks which have been publicly-traded since around 2005 with consistently available equity price data for each bank over the period January 1, 2006 to June 1, 2017. Our coverage of EA banks constitutes around 50 percent of that banking system’s assets.³⁷ All non-EA banks (allocated across the remaining regions) are drawn from a list of the top 100 (publicly-traded) global banks by assets size and is in line with Demirer and others (2017). Descriptive statistics indicate that average equity returns for all banks over the sample was slightly negative, likely reflecting the legacy of the global financial crisis (Appendix I, Table 1). In addition to applying the DY methodology to daily log returns, intra-day equity volatility series are also considered.³⁸ For each bank, this is computed as a function of the difference between maximum and minimum equity price, observed over a day; see Parkinson (1980).³⁹

73. Given the primary interest is investigating spillovers across regional banking systems, the general framework discussed above will need to be modified. Specifically, to facilitate estimation with data pooled across constituent banks within these systems. Appendix III details these modifications, and also discusses reasons to opt for such an estimation strategy given constraints posed by the nature of empirical investigation, and data availability.

Evolution of Net Spillovers

74. To give a sense of how spillovers have changed over time, the example of net spillovers between the euro area and U.S. banking systems is considered. Overall, the dynamics of net spillovers indicate that the euro area banking system shifts between periods of being a net recipient of spillovers and a transmitter vis-à-vis the U.S. banking system. Key events have influenced the net spillovers between these two banking systems including standard and unconventional monetary policy actions as well as episodes of acute financial distress.⁴⁰

Determinants of Net Spillovers

75. In this sub-section, the estimated (net) spillover measures are linked to their potential determinants. The primarily focus is on bank-specific financial soundness indicators as possible determinants which include capital buffers, profitability, asset quality, and a measure of short-term liquidity. Country-specific information on GDP developments is also conditioned upon. Table 12 summarizes how these determinants are proxied. Since balance sheet information tends to be recorded only at a quarterly or annual frequency, the selection of variables selected to proxy the aforementioned bank-level characteristics was guided by the consistent availability of quarterly data, covering the entire sample (2006 Q1–2017 Q2). To align data frequencies, the estimated daily (net) spillover series were converted to quarterly averages.

76. Descriptive statistics reveal some insightful findings regarding the bank-specific determinants. Profitability and asset quality, measured with ROA and the NPL ratio, are of particular interest. While average ROA is 0.15 percent, the median is 0.31 percent suggesting a profitability distribution with a long left tail populated by weaker banks (Appendix I, Table 2). Likewise, the average NPL ratio exceeds the median suggesting that asset quality issues are plaguing some banks to a much greater extent.

Table 12.

Determinants

Sources: Bloomberg, Datastream, Bankscope, SNL, and IMF staff. Notes: ROA is computed as ratio of trailing 12-month net income to average total assets. Country-specific information pertains to the country where bank is headquartered.

Table 12.

Determinants

Characteristics	Proxy variable	Label
Banks specific
Capital buffers	Ratio of Tier 1 capital to total assets	‘Tier 1’
Profitability	Return on assets	‘ROA’
Asset Quality	Ratio of nonperforming loans to total loans	‘NPL’
Short term liquidity	Ratio of cash and marketable securities to total liabilities	‘St. Liq.’
Country specific
GDP growth	Quarter-on-quarter growth rate	‘GDP’

Table 12.

Determinants

Characteristics	Proxy variable	Label
Banks specific
Capital buffers	Ratio of Tier 1 capital to total assets	‘Tier 1’
Profitability	Return on assets	‘ROA’
Asset Quality	Ratio of nonperforming loans to total loans	‘NPL’
Short term liquidity	Ratio of cash and marketable securities to total liabilities	‘St. Liq.’
Country specific
GDP growth	Quarter-on-quarter growth rate	‘GDP’

D. Pooled OLS Versus Quantile Regressions

Before proceeding to quantile regression analysis, this section discusses the results based on more familiar OLS regressions to get a broad sense of the link between net spillovers and their potential determinants. The expectation is that better financial soundness indicators would reduce the susceptibility to inward spillovers.

Pooled OLS Regression

77. There is an intuitive and statistically significant relationship between net spillovers and bank-specific characteristics. As a first pass, and to present some further stylized facts, OLS regressions are used to uncover the drivers of the net spillover measures.⁴¹ It is evident from results presented in Table 13 that most of the included banking system characteristics have a significant impact on net spillovers (both based on equity returns and volatility). The results are intuitive in that better capitalization, and for example, liquidity metrics are associated with higher net spillovers (which likely reflect, lower inward spillovers). Likewise, in line with expectations, the coefficient on the NPL ratio is negative and is statistically significant across the board. Real GDP growth appears to be statistically significant only the case of U.S. return and AE volatility spillovers.

Table 13.

Pooled OLS Regression Analysis

Note: The dependent variable, net spillover EA, refers to the dependent variable, net spillovers. (**) denotes significance at 5 percent level. Panel corrected standard errors computed using cross-sectional weights reported in parentheses.

Table 13.

Pooled OLS Regression Analysis

	Returns				Volatility
Net spillovers EA:	OE	US	AE	EM	OE	US	AE	EM
Tier 1 Capital Ratio	0.43**	0.11**	0.18**	0.12**	0.57**	0.30**	0.06**	0.08**
Tier 1 Capital Ratio	(0.64)	(0.03)	(0.03)	(0.03)	(0.06)	(0.04)	(0.03)	(0.02)
Return on Assets	0.29**	0.07	0.17**	0.21**	0.02	0.08	0.01	0.05
(ROA)	(0.11)	(0.06)	(0.07)	(0.06)	(0.12)	(0.08)	(0.06)	(0.04)
NPL ratio	−0.15**	−0.09**	−0.14**	−0.12**	−0.14**	−0.05**	−0.09**	−0.07**
NPL ratio	(0.02)	(0.02)	(0.01)	(0.01)	(0.02)	(0.02)	(0.01)	(0.00)
Short-term	0.15**	0.07**	0.08**	0.07**	0.18**	0.11**	0.05**	0.06**
Liquidity	(0.02)	(0.01)	(0.02)	(0.01)	(0.03)	(0.02)	(0.01)	(0.01)
GDP growth	0.02	0.03**	0.01	0.01	0.02	0.01	0.03**	−0.01
(country-specific)	(0.03)	(0.01)	(0.49)	(0.01)	(0.02)	(0.02)	(0.01)	(0.01)
Constant	−5.07**	−1.76**	2.46**	2.64**	−6.12**	−4.92**	1.55**	1.48**
Constant	(0.62)	(0.38)	(0.37)	(0.32)	(0.69)	(0.44)	(0.32)	(0.26)
Observations	1424	1424	1424	1424	1424	1424	1424	1424
R²	0.32	0.29	0.26	0.22	0.30	0.24	0.25	0.23

Table 13.

Pooled OLS Regression Analysis

	Returns				Volatility
Net spillovers EA:	OE	US	AE	EM	OE	US	AE	EM
Tier 1 Capital Ratio	0.43**	0.11**	0.18**	0.12**	0.57**	0.30**	0.06**	0.08**
Tier 1 Capital Ratio	(0.64)	(0.03)	(0.03)	(0.03)	(0.06)	(0.04)	(0.03)	(0.02)
Return on Assets	0.29**	0.07	0.17**	0.21**	0.02	0.08	0.01	0.05
(ROA)	(0.11)	(0.06)	(0.07)	(0.06)	(0.12)	(0.08)	(0.06)	(0.04)
NPL ratio	−0.15**	−0.09**	−0.14**	−0.12**	−0.14**	−0.05**	−0.09**	−0.07**
NPL ratio	(0.02)	(0.02)	(0.01)	(0.01)	(0.02)	(0.02)	(0.01)	(0.00)
Short-term	0.15**	0.07**	0.08**	0.07**	0.18**	0.11**	0.05**	0.06**
Liquidity	(0.02)	(0.01)	(0.02)	(0.01)	(0.03)	(0.02)	(0.01)	(0.01)
GDP growth	0.02	0.03**	0.01	0.01	0.02	0.01	0.03**	−0.01
(country-specific)	(0.03)	(0.01)	(0.49)	(0.01)	(0.02)	(0.02)	(0.01)	(0.01)
Constant	−5.07**	−1.76**	2.46**	2.64**	−6.12**	−4.92**	1.55**	1.48**
Constant	(0.62)	(0.38)	(0.37)	(0.32)	(0.69)	(0.44)	(0.32)	(0.26)
Observations	1424	1424	1424	1424	1424	1424	1424	1424
R²	0.32	0.29	0.26	0.22	0.30	0.24	0.25	0.23

Quantile Regression Analysis

78. Quantile regressions capture non-linear relationships between net spillovers and bank-specific characteristics. Recall that the OLS analysis indicated that the link between profitability (ROA) and net spillovers was not as strong relative to other bank-specific determinants, especially in the case of volatility spillovers. However, this result may be masking insightful non-linearities across different quantiles.

79. To this end, the OLS regressions discussed above are re-estimated, but using quantile regression analysis over a range of deciles. In this case, there would be 9 coefficients linking ROA to net spillovers (one coefficient for each decile), which are shown in Figure 13. This example includes net spillovers between the EA and two other region banking systems: Other Europe (OE) and the U.S. In both cases, the coefficients increase progressively from the lower to the upper deciles. At the same time, shaded bars denote statistically significant coefficients (at the 5 percent level), thereby highlighting the non-linear relationship between profitability and net spillovers— which is especially striking in the case of EA:U.S. spillovers. In sum, while a meaningful impact of profitability on the central tendency of net spillovers is evident in the case of EA:OE spillovers, this impact is much larger towards the right tail of the conditional distributions of net EA:OE and EA:US spillovers.

E. Conditional Spillover Distributions

This section sets the stage by discussing the baseline conditional distributions of net spillovers. Then various illustrative scenarios are considered whereby shifts in selected bank-specific determinants alter the shape of the net spillover distributions.

Baseline Distributions

80. The baseline conditional distributions of euro area net spillovers differ across regions. Figure 14 displays the baseline conditional probability density function of net spillovers, which is estimated by setting the values of all determinants to their sample averages. Four distributions are shown, corresponding to the euro area’s net spillovers vis-à-vis banking systems in Other Europe (OE), the United States (U.S.), Advanced Economies, excluding U.S. and those in Europe (AE), and emerging markets (EM). In terms of interpretation, moving rightwards (leftwards) along the horizontal axis corresponds to increasing magnitudes of outward (inward) spillovers. Given the similar probability mass on either side of zero, visual inspection of these densities suggests that the EA banking system is equally as likely to be a recipient of an inward spillovers from OE and U.S. banking systems, as it is to be a transmitter of outward spillovers to these systems. However, given its more leptokurtic shape, EA net spillovers vis-à-vis OE are more prone to tail risks. The EA appears much more likely to be a transmitter of outward spillovers to AE and EM banking systems, than being a recipient.

81. The conditional distributions facilitate quantitative assessments. The probability of EA net spillovers being less than or equal to zero is computed by integrating the area under the each of the baseline densities. These resulting cumulative probabilities, prob(net spillover EA: non-EA ≤ 0), are presented in Table 14, and are a convenient way of summarizing net spillovers across regions. The initial assessment regarding the likelihood of inward spillovers from each of the non-EA systems (Figure 14) broadly accords with this formal quantification (Table 14).

Table 14.

Probability of Inward Spillover—Baseline

(Percent)

Source: IMF staff estimates. Note: ‘netspill’ abbreviates net spillovers. prob(netspillEA: non-EA ≤ 0) computations are based on the cumulative distribution functions corresponding to a skewed-t form.

Table 14.

Probability of Inward Spillover—Baseline

(Percent)

prob(netspillEA:US ≤0)	prob(netspillEA:AE ≤0)	prob(netspill EA:EM ≤0)	prob(netspill EA:OE ≤0)
56.1	10.1	9.7	49.7

Table 14.

Probability of Inward Spillover—Baseline

(Percent)

prob(netspillEA:US ≤0)	prob(netspillEA:AE ≤0)	prob(netspill EA:EM ≤0)	prob(netspill EA:OE ≤0)
56.1	10.1	9.7	49.7

Scenario Analysis

82. The impact of changes to the bank-specific determinants on the baseline spillover densities is now examined. Both one and two standard deviation shocks relative to their average (baseline) values are used when generating the new (shocked) distributions. Note that for a particular determinant being shocked, a density’s central tendency may shift with minimal effect on tails or vice versa. In other instances there could be simultaneous shifts in the central tendency and tails of the distributions.

83. In general, the results suggest that stronger bank fundamentals reduce net spillovers from the rest of the world not only on average, but also in terms of tail risks. Densities conditional on shocked determinants are compared with their baselines in Figures 15–18. Consider the example of net spillovers between the EA and U.S. banking systems (Figure 15). Relative to the baseline, an increase in capital buffers appears to shift primarily the mode of the density, with both tails remaining anchored. As a result, these changes in the moments of the density translate into a decline in the probability of inward spillover to the EA banking system (Table 15). However, an increase in NPL ratio results in leftward shift in the central tendency accompanied with a retrenching of the right-tail thereby raising the likelihood of inward spillovers. In fact, on average, variations in the NPL ratio result in the largest changes in the probability of inward EA spillovers from the other banking systems. While the figures focus on rising NPL ratios, Table 16 considers negative shocks, which summarize the spillover implications owing to a lower NPL ratio (and complements Table 15).

Table 15.

Probability of Inward Spillover—Positive Shocks to Baseline

(In percent)

Source: IMF staff estimates. Note: ‘netspill’ abbreviates net spillovers. prob(netspillover EA: non-EA ≤ 0) computations are based on the cumulative distribution functions corresponding to a skewed-t form. ‘sd’ refers to standard deviations.

Table 15.

Probability of Inward Spillover—Positive Shocks to Baseline

(In percent)

prob(netspill EA:US ≤0)			prob(netspill EA:AE ≤0)		prob(netspill EA:EM ≤0)		prob (netspill EA:OE ≤0)
Baseline	56.1		10.1		9.7		49.7
Shock	+ 1 sd	+ 2 sd	+1 sd	+2 sd	+1 sd	+ 2 sd	+1 sd	+ 2 sd
Tier 1	49.1	43	8.8	7.6	8.7	8.3	40.5	32.6
ROA	54.1	52.2	9.1	8.3	8.6	7.7	47.7	46.1
NPL	72.1	87.6	13.2	21.1	12.7	18.8	63.2	74.4
St. Liq.	51	44.1	7.6	5.7	6.9	4.8	44.3	37.9

Table 15.

Probability of Inward Spillover—Positive Shocks to Baseline

(In percent)

prob(netspill EA:US ≤0)			prob(netspill EA:AE ≤0)		prob(netspill EA:EM ≤0)		prob (netspill EA:OE ≤0)
Baseline	56.1		10.1		9.7		49.7
Shock	+ 1 sd	+ 2 sd	+1 sd	+2 sd	+1 sd	+ 2 sd	+1 sd	+ 2 sd
Tier 1	49.1	43	8.8	7.6	8.7	8.3	40.5	32.6
ROA	54.1	52.2	9.1	8.3	8.6	7.7	47.7	46.1
NPL	72.1	87.6	13.2	21.1	12.7	18.8	63.2	74.4
St. Liq.	51	44.1	7.6	5.7	6.9	4.8	44.3	37.9

Table 16.

Probability of Inward Spillover—Negative Shocks to Baseline

(Percent)

Source: IMF staff estimates. Note: ‘netspill’ abbreviates net spillovers. p(netspillover EA: non-EA ≤ 0) computations are based on the cumulative distribution functions corresponding to a skewed-t form. ‘sd’ refers to standard deviations.

Table 16.

Probability of Inward Spillover—Negative Shocks to Baseline

(Percent)

	p(netspill EA:US ≤0)		p(netspill EA:AE ≤0)		p(netspill EA:EM ≤0)		p(netspill EA:OE ≤0)
Baseline	56.1		10.1		9.7		49.7
Shock	−1 sd	− 2 sd	−1 sd	−2 sd	−1 sd	−2 sd	−1 sd	−2 sd
Tier 1	63.2	69.5	11.2	13.2	10.4	11.4	60.8	71.8
ROA	58.1	60.4	10.7	10.8	10.5	11.1	52	54.7
NPL	43.5	34.6	7.9	6.8	7.4	6.7	39.1	32.2
St. Liq.	56	63.2	12.7	15.8	13	16.4	54.8	59.6

Table 16.

Probability of Inward Spillover—Negative Shocks to Baseline

(Percent)

	p(netspill EA:US ≤0)		p(netspill EA:AE ≤0)		p(netspill EA:EM ≤0)		p(netspill EA:OE ≤0)
Baseline	56.1		10.1		9.7		49.7
Shock	−1 sd	− 2 sd	−1 sd	−2 sd	−1 sd	−2 sd	−1 sd	−2 sd
Tier 1	63.2	69.5	11.2	13.2	10.4	11.4	60.8	71.8
ROA	58.1	60.4	10.7	10.8	10.5	11.1	52	54.7
NPL	43.5	34.6	7.9	6.8	7.4	6.7	39.1	32.2
St. Liq.	56	63.2	12.7	15.8	13	16.4	54.8	59.6

84. The reduction of the probability of inward spillovers differs across regions and the bank-specific shocks under consideration. For instance, higher profitability leads to a similar decline in the probability of inward EA spillovers vis-à-vis the U.S. and OE banking systems (Tables 15–16). However, relative to profitability, greater capitalization results in even greater decreases, especially in the context of EA:OE spillovers. This said, a decline in the NPL ratio results in the greatest decline in the probability of inward spillovers for all banking systems.

85. Shocks can also affect the spillover distributions in an asymmetric manner given the shape of the baseline and shocked distributions. Such differences are greatest in the case of NPLs (when comparing baseline and shocked probabilities): although a two standard deviation increase in the ratio increases the probability of inward spillovers by about 32 percentage points (Table 15), the analogous decrease results in a decline of about 21 percentage points (Table 16). This result underscores how much a deterioration in key bank fundamentals can heightened vulnerabilities to spillovers from other banking systems.

Pre- and Post-Sample Analysis

86. The relationship between spillovers and bank fundamentals has evolved over the past decade. The models are re-estimated using two sub-samples: (1) 2006 Q1–2012 Q3, and (2) 2012 Q4–2017 Q2. The results are shown in Figure 19, and documented in Table 17. Stronger bank-specific fundamentals, such as better asset quality, in recent years appear to reduce the probability of inwards spillovers to a greater extent relative to earlier periods.

Table 17.

Probability of Inward Spillover—Comparing Pre-, and Post-Sample

(Percent)

Source: IMF staff estimates. Note: ‘netspill’abbreviates net spillovers. p(netspill EA: non-EA ≤ 0) computations are based on the cumulative distribution functions corresponding to a skewed-t form. ‘sd’ refers to standard deviations.

Table 17.

Probability of Inward Spillover—Comparing Pre-, and Post-Sample

(Percent)

	p(netspill EA:US ≤0)		p(netspill EA:AE ≤0)		p(netspill EA:EM ≤0)		p(netspill EA:OE ≤0)
Sample split	Pre	Post	Pre	Post	Pre	Post	Pre	Post
Baseline	54.0	55.0	9.0	9.1	8.1	10.0	56.1	43.2
Shock	+2 sd	+ 2 sd	+2sd	+ 2 sd	+2 sd	+2 sd	+2 sd	+ 2 sd
Tier 1 Ratio	43.2	53.1	6.4	8.0	4.1	9.1	50.0	38.1
ROA	51.3	47.2	8.1	8.1	7.0	7.0	54.1	31.1
NPL	58.4	83.1	12.3	26.2	11.2	28.0	66.2	76.4
St. Liq.	48.0	42.1	7.1	6.0	7.0	5.0	50.3	22.7

Table 17.

Probability of Inward Spillover—Comparing Pre-, and Post-Sample

(Percent)

	p(netspill EA:US ≤0)		p(netspill EA:AE ≤0)		p(netspill EA:EM ≤0)		p(netspill EA:OE ≤0)
Sample split	Pre	Post	Pre	Post	Pre	Post	Pre	Post
Baseline	54.0	55.0	9.0	9.1	8.1	10.0	56.1	43.2
Shock	+2 sd	+ 2 sd	+2sd	+ 2 sd	+2 sd	+2 sd	+2 sd	+ 2 sd
Tier 1 Ratio	43.2	53.1	6.4	8.0	4.1	9.1	50.0	38.1
ROA	51.3	47.2	8.1	8.1	7.0	7.0	54.1	31.1
NPL	58.4	83.1	12.3	26.2	11.2	28.0	66.2	76.4
St. Liq.	48.0	42.1	7.1	6.0	7.0	5.0	50.3	22.7

Improved Fundamentals and Positive Externalities

87. The impact of progressively stronger fundamentals on the probability of outward spillovers is now investigated. Figure 20 demonstrates the limiting behavior of conditional net spillover distributions due to progressively higher EA capitalization ratios. Bilateral spillovers between the euro area and U.S. bank systems is used as an illustration. Overall, the findings suggest that stronger fundamentals can improve a banking system’s resilience to inward spillovers without necessarily aggravating outward spillovers. This is evident in the variance of the distributions progressively tightening around a particular point of the distributions’ support (that is, the central tendency stops shifting rightwards). Given that an upper bound on magnitude of outward spillover exists (in the limit) strong euro area fundamentals appear to enhance the stability of other regions. Broadly similar results are found for other bank fundamentals (for instance, NPL ratios) and the other three regions.

Spillovers Within the Euro Area

88. Although not the focus of this chapter, the analysis also considered within euro area banking spillovers. Recall that the euro area was taken as a single banking system in the discussions above. Now, each of the 10 euro area countries in the sample correspond to a banking system. While the results are suppressed for brevity, spillovers across these 10 countries was analyzed using the same approach outlined above. Overall, consistent with the results discussed above, enhanced banking soundness in a euro area country reduces the likelihood of inward spillovers from the rest of the euro area, both in terms of central tendencies, but also in terms of tail risks.

F. Conclusions and Policy Implications

89. This chapter proposes a novel framework to quantify and appraise risks associated with euro area banking spillovers. The framework derives probability distributions of net euro area spillovers conditional on selected determinants vis-à-vis other global banking systems. The findings suggest that stronger bank fundamentals (lower NPL ratios, greater profitability, and higher capitalization levels) reduce net spillovers to the euro area banking system from the rest of the world not only on average, but also in terms of tail risks. Moreover, such effects appear to have strengthen in recent years. Increasingly stronger fundamentals can enhance the euro area banking system’s resilience to inward spillovers without necessarily aggravating outward spillovers.

90. Even though EA banking system soundness indicators have been improving, addressing certain structural challenges could further reduce vulnerability to inward spillovers. EA banks’ capital ratios have risen on average, capital quality has improved, and funding has become more stable as banks are increasingly relying on deposits. At the same time, despite improvements, progress in reducing NPLs remains uneven and bank profitability remains generally low. Therefore, some banks in particular, should take advantage of the current upswing to resolutely address their NPL stocks. In addition, greater cost efficiency (via digitization, for example) and a tailored approach to revamping business models could support the profitability prospects of many banks. Further progress on both fronts would reduce the likelihood of inwards spillovers to the EA banking system. Relatedly, the interconnectedness framework—already very sophisticated—could be further enhanced by developing tools that help quantify the tail risks associated with inward spillovers.

References

Adrian, T., Markus K, B., (2016), “CoVAR,” American Economic Review 106(7), pp. 1705–1741.
- Search Google Scholar
- Export Citation
Adrian, T., Boyarchenko, N., Giannone, D., (2016), “Vulnerable Growth.” FRB of NY Staff Report, No. 794.
- Search Google Scholar
- Export Citation
Azzalini, A. and Capitanio, A. (2003), “Distributions generated by perturbation of symmetry with emphasis on a multivariate skew t-distribution.” Journal of the Royal Statistical Society: Series B (Statistical Methodology), 65, pp. 367–389.
- Search Google Scholar
- Export Citation
Boudt, K., Carl, P., Peterson, B. G., (2012), “Asset Allocation with Conditional Value-at-Risk Budgets,” Journal of Risk 15(3), pp. 39–68.
- Search Google Scholar
- Export Citation
Demirer, M., Diebold, F.X., Liu, L., Yilmaz, K. (2017), “Estimating Global Bank Network Connectedness,” Manuscript, University of Pennsylvania.
- Search Google Scholar
- Export Citation
Diebold, F. X., Yilmaz, K., (2009), “Measuring Financial Asset Return and Volatility Spillovers, with Application to Global Equity Markets,” Economic Journal (119), pp. 158–171.
- Search Google Scholar
- Export Citation
Diebold, F. X., Yilmaz, K., (2014), “On the Network Topology of Variance Decompositions: Measuring the Connectedness of Financial Firms,” Journal of Econometrics, vol. 182(1), pp. 119–134.
- Search Google Scholar
- Export Citation
Diebold, F. X., Yilmaz, K., (2015), “Trans-Atlantic Equity Volatility Connectedness: U.S. and European Financial Institutions, 2004–2014,” Journal of Financial Econometrics, Society for Financial Econometrics, vol. 14(1), pp. 81–127.
- Search Google Scholar
- Export Citation
Engle, R., Ng, V. K., Rothschild, M., (1990), “Asset pricing with a factor-arch covariance structure: Empirical estimates for treasury bills,” Journal of Econometrics, 45(1–2), pp. 213–237.
- Search Google Scholar
- Export Citation
Fernández, C., Steel, M. F. J., (2012), “On Bayesian Modeling of Fat Tails and Skewness,” Journal of the American Statistical Association, 93:441, pp. 359–371,
- Search Google Scholar
- Export Citation
Forbes, K. J. and Rigobon, R. (2002), “No Contagion, Only Interdependence: Measuring Stock Market Comovements,” The Journal of Finance, 57, pp. 2223–2261.
- Search Google Scholar
- Export Citation
GFSR (April 2016), “The Growing Importance of Financial Spillovers from Emerging Market Economies” GFSR Report, Chapter 2.
- Search Google Scholar
- Export Citation
GFSR (October 2017), “Financial Conditions and Growth at Risk” GFSR Report, Chapter 3.
- Search Google Scholar
- Export Citation
Giot, P., Laurent, S. (2003), “Value-at-risk for long and short trading positions,” Journal of Applied Economics, 18, pp. 641–663.
- Search Google Scholar
- Export Citation
Hansen, B., (1994), “Autoregressive Conditional Density Estimation,” International Economic Review, 35(3), pp. 705–30.
- Search Google Scholar
- Export Citation
Koenker, r., Basset, G., (1978), “Regression Quantiles,” Econometrica, 46, pp. 33–50.
- Search Google Scholar
- Export Citation
Koenker, R., Ng, P., (2005), “Inequality Constrained Quantile Regression” Sankhyā: The Indian Journal of Statistics (2003–2007), 67(2), pp. 418–440.
- Search Google Scholar
- Export Citation
Koop, G., Pesaran, M. H., Potter, S. M., (1996), “Impulse response analysis in nonlinear multivariate models,” Journal of Econometrics 74, pp. 119–147.
- Search Google Scholar
- Export Citation
Lambert, P., Laurent, S., (2002), “Modeling skewness dynamics in series of financial data using skewed location-scale distributions,” Working Paper, Universite’ Catholique de Louvain and Universite’ de Liege.
- Search Google Scholar
- Export Citation
Malik, S., Xu, T., (2017), “Interconnectedness of Global Systemically-Important Banks and Insurers,” IMF Working Paper, 17/210.
- Search Google Scholar
- Export Citation
Parkinson, M., (1980), “The Extreme Value Method for Estimating the Variance of the Rate of Return,” The Journal of Business, 53(1), pp. 61–65.
- Search Google Scholar
- Export Citation
Pesaran, H. H., Shin, Y., (1998), “Generalized impulse response analysis in linear multivariate models,” Economics Letters, 58(1), pp. 17–29,
- Search Google Scholar
- Export Citation
Sims, C. A., (1980), “Macroeconomics and Reality,” Econometrica, 48, pp. 1–48.
- Search Google Scholar
- Export Citation
Tibshirani, R. (1996), “Regression Shrinkage and Selection via the Lasso,” Journal of the Royal Statistical Society. Series B (Methodological), 58(1), pp. 267–288.
- Search Google Scholar
- Export Citation
Zou, H. and Hastie, T. (2005), “Regularization and variable selection via the elastic net,” Journal of the Royal Statistical Society: Series B (Statistical Methodology), 67, pp. 301–320.
- Search Google Scholar
- Export Citation

Appendix I. Market and Balance Sheet Indicators

Appendix I Table 1.

Market Based Indicators: Descriptive Statistics of Banks in the Sample ¹/

(2006Q1–2017Q2, in percent)

Sources: Bloomberg, and IMF staff calculations.

^¹/Sample includes top traded 93 banks by asset size

^²/Log returns based on the end of day equity price

^³/Intraday equity volatility, annualized.

Appendix I Table 1.

Market Based Indicators: Descriptive Statistics of Banks in the Sample ¹/

(2006Q1–2017Q2, in percent)

	Equity returns ²/	Equity volatility ³/
Mean	−0.018	34.65
Median	0.000	25.67
Standard Deviation	2.81	33.53
Observations	280953	280953

Sources: Bloomberg, and IMF staff calculations.

^¹/Sample includes top traded 93 banks by asset size

^²/Log returns based on the end of day equity price

^³/Intraday equity volatility, annualized.

Appendix I Table 1.

Market Based Indicators: Descriptive Statistics of Banks in the Sample ¹/

(2006Q1–2017Q2, in percent)

	Equity returns ²/	Equity volatility ³/
Mean	−0.018	34.65
Median	0.000	25.67
Standard Deviation	2.81	33.53
Observations	280953	280953

Sources: Bloomberg, and IMF staff calculations.

^¹/Sample includes top traded 93 banks by asset size

^²/Log returns based on the end of day equity price

^³/Intraday equity volatility, annualized.

Appendix I Table 2.

Balance Sheet Determinants: Descriptive Statistics of Banks in the Sample ¹/

(2006Q1–2017Q2, in percent)

Sources: Bloomberg, and IMF staff calculations.

^¹/Sample includes top traded 93 banks by asset size

^²/Ratio of Tier 1 capital to total assets

^³/Ratio of trailing 12-month income to average total assets

^⁴/Ratio of nonperforming loans to total loans

^⁵/Ratio of cash and marketable securities to total liabilities.

Appendix I Table 2.

Balance Sheet Determinants: Descriptive Statistics of Banks in the Sample ¹/

(2006Q1–2017Q2, in percent)

	Tier 1 ratio ²/	Return on assets ³/	NPL ratio ⁴/	Liquidity ⁵/
Mean	10.80	0.15	7.74	9.79
Median	11.00	0.31	5.14	9.34
Standard Deviation	2.90	1.36	8.19	5.60
Observations	1242	1242	1242	1242

Sources: Bloomberg, and IMF staff calculations.

^¹/Sample includes top traded 93 banks by asset size

^²/Ratio of Tier 1 capital to total assets

^³/Ratio of trailing 12-month income to average total assets

^⁴/Ratio of nonperforming loans to total loans

^⁵/Ratio of cash and marketable securities to total liabilities.

Appendix I Table 2.

Balance Sheet Determinants: Descriptive Statistics of Banks in the Sample ¹/

(2006Q1–2017Q2, in percent)

	Tier 1 ratio ²/	Return on assets ³/	NPL ratio ⁴/	Liquidity ⁵/
Mean	10.80	0.15	7.74	9.79
Median	11.00	0.31	5.14	9.34
Standard Deviation	2.90	1.36	8.19	5.60
Observations	1242	1242	1242	1242

Sources: Bloomberg, and IMF staff calculations.

^¹/Sample includes top traded 93 banks by asset size

^²/Ratio of Tier 1 capital to total assets

^³/Ratio of trailing 12-month income to average total assets

^⁴/Ratio of nonperforming loans to total loans

^⁵/Ratio of cash and marketable securities to total liabilities.

Appendix II. Sample Description

Appendix II Table 1.

Sample of Banks and System Groupings

Source: IMF Staff, Diebold-Yilmaz (2015). Note: Advanced Economies excludes United States.

Appendix II Table 1.

Sample of Banks and System Groupings

Euro Area	Other Europe	United States	Advanced Economies	Emerging Markets
Credit Agricole	Barclays	Bank of America	Mitsubishi UFJ Financial Group	Ping An Bank Co
BNP Paribas	Credit Suisse Group	Bank of New York Mellon	Sumitomo Mitsui Financial Group	Shanghai Pudong Development
Deutsche Bank	HSBC Holdings	Citigroup	Mizuho Financial Group	Bank Co
Societe Generale	Nordea Bank	Goldman Sachs Group	Industrial Bank of Korea	Huaxia Bank Co
ING	Royal Bank of Scotland	JPMorgan Chase & Co	Shinhan Financial Group	China Minsheng Banking Corp
Banco Santander	Group	Morgan Stanley	Resona Holdings	China Merchants Bank Co
UniCredit	Standard Chartered	State Street Corp	Chiba Bank/The	Banco Bradesco
Allied Irish Banks	UBS Group	Wells Fargo & Co	Shizuoka Bank/The	Bank of Baroda
Alpha Bank	Danske Bank	American Express	Hokuhoku Financial Group	CIMB Group Holdings Bhd
Banco Bilbao Vizcaya Argentaría	DNB	BB&T Corp	Nomura Holding	Turkiye Is Bankasi
Banco Comercial Portugues	Lloyds Banking Group	Capital One Financial Corp	Australia & New Zealand Banking Group	Itau Unibanco Holding
Bank of Ireland	Skandinaviska Enskilda	Fifth Third Bancorp	Bank of Montreal	Malayan Banking Bhd
Bankinter	Banken	The PNC Financial Services	The Bank of Nova Scotia	Sberbank of Russia PJSC
BPER Banca	Svenska Handelsbanken	Group Inc	Commonwealth Bank of Australia	State Bank of India
Commerzbank	Swedbank	Regions Financial Corp	Canadian Imperial Bank of Commerce	Standard Bank Group
Credit Industriel et Commercial		SunTrust Banks	DBS Group Holdings
Dexia		US Bancorp	Maquarie Group
Erste Group Bank			National Bank of Canada
National Bank of Greece			National Australia Bank
Intesa Sanpaolo			Royal Bank of Canada
KBC Group			The Toronto-Dominion Bank
Natixis			United Overseas Bank
Mediobanca			Westpac Bankingrp
Banco de Sabadell
Piraeus Bank
Unione di Banche Italiane
Unipol Gruppo Finanziaro

Source: IMF Staff, Diebold-Yilmaz (2015). Note: Advanced Economies excludes United States.

Appendix II Table 1.

Sample of Banks and System Groupings

Euro Area	Other Europe	United States	Advanced Economies	Emerging Markets
Credit Agricole	Barclays	Bank of America	Mitsubishi UFJ Financial Group	Ping An Bank Co
BNP Paribas	Credit Suisse Group	Bank of New York Mellon	Sumitomo Mitsui Financial Group	Shanghai Pudong Development
Deutsche Bank	HSBC Holdings	Citigroup	Mizuho Financial Group	Bank Co
Societe Generale	Nordea Bank	Goldman Sachs Group	Industrial Bank of Korea	Huaxia Bank Co
ING	Royal Bank of Scotland	JPMorgan Chase & Co	Shinhan Financial Group	China Minsheng Banking Corp
Banco Santander	Group	Morgan Stanley	Resona Holdings	China Merchants Bank Co
UniCredit	Standard Chartered	State Street Corp	Chiba Bank/The	Banco Bradesco
Allied Irish Banks	UBS Group	Wells Fargo & Co	Shizuoka Bank/The	Bank of Baroda
Alpha Bank	Danske Bank	American Express	Hokuhoku Financial Group	CIMB Group Holdings Bhd
Banco Bilbao Vizcaya Argentaría	DNB	BB&T Corp	Nomura Holding	Turkiye Is Bankasi
Banco Comercial Portugues	Lloyds Banking Group	Capital One Financial Corp	Australia & New Zealand Banking Group	Itau Unibanco Holding
Bank of Ireland	Skandinaviska Enskilda	Fifth Third Bancorp	Bank of Montreal	Malayan Banking Bhd
Bankinter	Banken	The PNC Financial Services	The Bank of Nova Scotia	Sberbank of Russia PJSC
BPER Banca	Svenska Handelsbanken	Group Inc	Commonwealth Bank of Australia	State Bank of India
Commerzbank	Swedbank	Regions Financial Corp	Canadian Imperial Bank of Commerce	Standard Bank Group
Credit Industriel et Commercial		SunTrust Banks	DBS Group Holdings
Dexia		US Bancorp	Maquarie Group
Erste Group Bank			National Bank of Canada
National Bank of Greece			National Australia Bank
Intesa Sanpaolo			Royal Bank of Canada
KBC Group			The Toronto-Dominion Bank
Natixis			United Overseas Bank
Mediobanca			Westpac Bankingrp
Banco de Sabadell
Piraeus Bank
Unione di Banche Italiane
Unipol Gruppo Finanziaro

Source: IMF Staff, Diebold-Yilmaz (2015). Note: Advanced Economies excludes United States.

Appendix III. Pooled Estimation Strategy

The general framework discussed above needs to be modified given that the primary interest is investigating spillovers across regional banking systems. From the perspective of undertaking bank-by-bank estimation of equation (2) (as currently formulated), a time series dimension of T = 44 quarters would prove rather restrictive in terms of allowing accurate pinning down of parameters; especially in a quantile regression setting aimed at informing higher order moments. A case can be made to pooling time series information across individual banks by estimating a stacked version of equation (2), i.e., stacking the cross-section of N banks, say, providing and estimation sample of N × T. The advantage of pooling strategy is that whilst circumventing the practical issue of too few degrees of freedom, estimated parameters will correspond to an average across the cross section of banks, which aligns with the objective of analyzing results at the level of a system’s average bank.

The definition of net spillovers needs to accordingly be adjusted. Given the specific focus of analyzing EA spillovers, information will be pooled over the cross-section of EA banks. To maintain consistency with regards to analyzing spillovers at the average EA/non-EA level within a pooled setting, expression (1) for net spillover will need to be modified. By way of example, suppose the non-EA system labeled AE consists of a set of banks indexed by s, where, s = 1,…,S. In this setting, $C_{i \to \bar{s}}^{H} (= \frac{1}{s} Σ_{s} C_{i \to s}^{H})$ denotes spillover from EA bank i, to the average AE bank. Conversely, $C_{i \leftarrow \bar{s}}^{H} (= \frac{1}{s} Σ_{s} C_{i \leftarrow s}^{H})$ corresponds to spillover to EA bank i, from the average AE bank. Therefore, the net spillover (for a single point in time) can be cast as:

\begin{matrix} n e t s p i l l o v e r_{i \leftrightarrow \bar{s}} = C_{i \to \bar{s}}^{H} - C_{i \leftarrow \bar{s}}^{H} & (5) \end{matrix}

Letting the number of EA banks in the total sample be given by N_EA such that i = 1, … N_EA with quarters t, such that t = 1,,…,T, the time series of net spillovers for each EA bank i vis-à-vis the average AE bank can then be stacked in order to estimate,

\begin{matrix} n e t s p i l l o v e r_{i \leftrightarrow \bar{s}, t}^{q} = α + B^{q} B a n k_{i, t} + G^{q} M a c r o_{i, t} + ε_{i, t}^{q} . & (6) \end{matrix}

Net spillovers of the EA system vis-s-vis the others is labeled as netspillover_EA: non-EA, where non-EA = {OE, AE, U.S., EM}.

This chapter was prepared by Selim Elekdag, Sheheryar Malik, and Tadeusz Galeza, all Monetary and Capital Markets Department, IMF.

The bulk of earlier work measures financial system interconnectedness without exploring the drivers of these dynamics. For example, Diebold and Yilmaz (2009) propose a simple quantitative measure of interdependence called a spillover index. They build on a vast literature including, for example, Engle (1990) and Engle and Kelly (2012). Demirer and others (2017) focus on global banks, and illustrate connectedness via network maps and summarize dynamics using rolling windows. However, except for IMF (2016) and Malik and Xu (2017), the literature stops short of explaining the potential factors influencing these patterns. Moreover, all papers focus on average spillover dynamics and do not consider higher moments or distributions.

The focus on net spillovers is to help facilitate the analysis and to present a summary metric, and not necessarily because net spillovers are more relevant to systemic risk analysis than gross spillovers.

Following Demirer and others (2017), the elastic net estimator (Zou and Hastie, 2005) is used to estimate the highdimensional VAR. This blends shrinkage and selection to recover degrees of freedom, to deal with the “curse of dimensionality.” Essentially, the elastic net estimator blends the Lasso (Tibshirani, 1996) and ridge regression. For the error variance decomposition, the Generalized Variance Decomposition (GVD) (Koop, Pesaran and Potter, 1996 and Pesaran and Shin, 1998) is applied. Compared to the Cholesky decomposition proposed by Sims (1980) and related identification strategies, GVD is invariant to the ordering of variables, which offers more flexibility in modeling strategy without making any a priori assumption on the sequence of responses.

Alternative specifications for the skewed t-distribution are present in literature, e.g., as put forth inter alia by Hansen (1994) and Azzalini and Capitanio (2003). These are essentially equivalent given a (nonlinear) transformation of the skewness parameter.

It is important to note that market data is not available even for large euro area some banks (and therefore the sample here differs from that used in the balance sheet-based interconnected analysis, for instance). More generally, there are some caveats to using market-based data that need to be recognized. For example, (thinly-traded) markets can underreact in tranquil times and overreact during episodes of stress (possibly reflecting, for instance, herding behavior), and therefore may not fully capture the build-up and unwinding of certain vulnerabilities. Bank returns may also be impacted by board market developments unrelated to the performance of the bank under consideration. Note also that there is most likely a loss of information when transforming the higher-frequency market data to lower frequencies (via averaging) which is required as the balanced sheet data is only available on a quarterly basis.

Accounting for various global factors as (exogenous) controls, such as measures of global interest rates as well as stock and bond market volatility metrics (VIX and the MOVE) does not materially change the spillover estimates.

To control for the differences in trading hours due to time zones, average two-day log returns for equity prices in local currency are computed (see, for example, Forbes and Rigobon, 2002, and GFSR, April 2016b). In order to deal with holidays and missing observations, a day is removed if more than half of the entities have missing data; remaining missing observations are then interpolated.

A 250-day window was used to calculate the spillovers, and a length of 150 days yields comparable results.

Although these OLS regressions are not the main focus of the chapter, the results were robust when other variable combinations were considered, for instance, log(asset) to control for size or the growth rates of the other bank-specific characteristics (e.g., tier 1 capital ratios).

Other Publishers

Save
Cite
Email this content

Share Link

Copy this link, or click below to email it to a friend
Email this content
or copy the link directly:

https://www.elibrary.imf.org/view/journals/002/2018/231/article-A003-en.xml
The link was not copied. Your current browser may not support copying via this button.

Link copied successfully

Collapse
Expand

Euro Area Policies: Financial Sector Assessment Program-Technical Note-Systemic Risk Analysis

Author:

International Monetary Fund. Monetary and Capital Markets Department

Volume/Issue:: Volume 2018: Issue 231

Publisher:: International Monetary Fund

ISBN:: 9781484369586

ISSN:: 1934-7685

Pages:: 94

DOI:: https://doi.org/10.5089/9781484369586.002

View raw image

Figure 12.

Net Spillovers 1/

(Index)

View raw image

Figure 13.

Non-Linear Impact of Profitability on Spillovers Across Percentiles

View raw image

Figure 14.

Baseline Probability Density Functions

View raw image

Figure 15.

Shocks to Baseline Probability Density Functions, EA:US

View raw image

Figure 16.

Shocks to Baseline Probability Density Functions, EA:AE

View raw image

Figure 17.

Shocks to Baseline Probability Density Functions, EA:EM

View raw image

Figure 18.

Shocks to Baseline Probability Density Functions, EA:OE

View raw image

Figure 19.

Shocks to Baseline Probability Density Functions, EA:AE Pre-, and Post- Sample Comparison

View raw image

Figure 20.

Impact of Improved Fundamentals

Foresee

Appendix II Table 1.

Sample of Banks and System Groupings

Euro Area	Other Europe	United States	Advanced Economies	Emerging Markets
Credit Agricole	Barclays	Bank of America	Mitsubishi UFJ Financial Group	Ping An Bank Co
BNP Paribas	Credit Suisse Group	Bank of New York Mellon	Sumitomo Mitsui Financial Group	Shanghai Pudong Development
Deutsche Bank	HSBC Holdings	Citigroup	Mizuho Financial Group	Bank Co
Societe Generale	Nordea Bank	Goldman Sachs Group	Industrial Bank of Korea	Huaxia Bank Co
ING	Royal Bank of Scotland	JPMorgan Chase & Co	Shinhan Financial Group	China Minsheng Banking Corp
Banco Santander	Group	Morgan Stanley	Resona Holdings	China Merchants Bank Co
UniCredit	Standard Chartered	State Street Corp	Chiba Bank/The	Banco Bradesco
Allied Irish Banks	UBS Group	Wells Fargo & Co	Shizuoka Bank/The	Bank of Baroda
Alpha Bank	Danske Bank	American Express	Hokuhoku Financial Group	CIMB Group Holdings Bhd
Banco Bilbao Vizcaya Argentaría	DNB	BB&T Corp	Nomura Holding	Turkiye Is Bankasi
Banco Comercial Portugues	Lloyds Banking Group	Capital One Financial Corp	Australia & New Zealand Banking Group	Itau Unibanco Holding
Bank of Ireland	Skandinaviska Enskilda	Fifth Third Bancorp	Bank of Montreal	Malayan Banking Bhd
Bankinter	Banken	The PNC Financial Services	The Bank of Nova Scotia	Sberbank of Russia PJSC
BPER Banca	Svenska Handelsbanken	Group Inc	Commonwealth Bank of Australia	State Bank of India
Commerzbank	Swedbank	Regions Financial Corp	Canadian Imperial Bank of Commerce	Standard Bank Group
Credit Industriel et Commercial		SunTrust Banks	DBS Group Holdings
Dexia		US Bancorp	Maquarie Group
Erste Group Bank			National Bank of Canada
National Bank of Greece			National Australia Bank
Intesa Sanpaolo			Royal Bank of Canada
KBC Group			The Toronto-Dominion Bank
Natixis			United Overseas Bank
Mediobanca			Westpac Bankingrp
Banco de Sabadell
Piraeus Bank
Unione di Banche Italiane
Unipol Gruppo Finanziaro

Source: IMF Staff, Diebold-Yilmaz (2015). Note: Advanced Economies excludes United States.

Topics

Countries

Series

About

Resources

Abstract

Market-Based Interconnectedness Analysis32

A. Introduction

B. General Framework: An Overview

Measuring Spillovers

Generating Conditional Distributions

C. Data and Stylized Facts

Market-based Data

Evolution of Net Spillovers

Determinants of Net Spillovers

D. Pooled OLS Versus Quantile Regressions

Pooled OLS Regression

Quantile Regression Analysis

E. Conditional Spillover Distributions

Baseline Distributions

Scenario Analysis

Pre- and Post-Sample Analysis

Improved Fundamentals and Positive Externalities

Spillovers Within the Euro Area

F. Conclusions and Policy Implications

References

Appendix I. Market and Balance Sheet Indicators

Appendix II. Sample Description

Appendix III. Pooled Estimation Strategy

Same Series

Other IMF Content

Other Publishers

Asian Development Bank

Inter-American Development Bank

Nordic Council of Ministers

The World Bank

Share Link

Table of Contents

Headings

Figures

Metrics

Foresee

Market-Based Interconnectedness Analysis³²