The multi-country transmission of sovereign and banking risk: a spatial vector autoregressive approach

ABSTRACT

This paper develops a spatial vector autoregressive (SpVAR) model to investigate the transmission of sovereign, banking and corporate default risks among 11 Eurozone countries for the period January 2008–December 2013. The results show that a significant proportion of default risk variation is explained by foreign shocks. However, the cross-border sovereign–bank nexus is statistically significant, but economically moderate. Among the three sectors, shocks to the banking sector play the most critical role. On average, for the 11 countries, a foreign banking shock can explain 7%, 23% and 18% of the forecast error variance of changes in sovereign, banking and corporate credit default swap spreads respectively.

KEYWORDS:

• sovereign credit default swap spreads
• banking credit default swap spreads
• corporate credit default swap spreads
• cross-border spillover
• spatial vector autoregressive model

JEL:

• C31
• C33
• F3

ACKNOWLEDGMENTS

The author thank three anonymous referees, the editor, Paul Elhorst, Bernd Schwaab, Gabriel Lee, Rolf Tschernig, and the participants at the Lunch Seminar at the University of Regensburg, The European Sovereign Debt Crisis Symposium, International Finance and Banking Society Conference, for useful suggestions. The author alone is responsible for any errors. This paper was complemented when the author worked at the University of Cambridge.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the author.

ORCID

Bing Zhu http://orcid.org/0000-0001-9757-3582

Notes

1 The corporates in this paper always refer to non-financial private corporates.

2 This paper follows the spatial econometric literature (Baltagi, Fingleton, & Pirotte, 2014; Cliff & Ord, 1973; Kelejian & Prucha, 1999, 2010; Lee, 2004) and uses known spatial weighting matrices to identify the pattern of dependence between cross-sectional and cross-border units.

3 For example, by means of a global VAR model, Gray et al. (2013) calculate the interdependencies of default risk among banks, sovereigns and corporates for 15 European countries and the United States. Alter and Beyer (2014) create a contagion index that measures the intensity of spillovers of sovereign and banking default risk among Eurozone countries. They find that the spillover intensities of Greece, Portugal and Ireland decreased at the beginning of 2012, which implies that European Central Bank (ECB) interventions successfully reduced systemic risk. Billio et al. (2014) study the transmission of credit risks of major European, US and Japanese banks using contingent claims analyses and Granger causality analyses. They conclude that banks became important sources of credit risk before and during the 2007–09 global financial crisis. After 2009, sovereign shock played a more important role. Based on the mixed cross-section GVAR model, the pioneer works of Gross and Kok (2013) and Gross, Kok, and Yochowski (2016) study the propagation of bank capital shocks to the economy. In addition to the GVAR model, SpVAR also takes the similar form as the infinite dimensional vector autoregressive model proposed by Chudik and Pesaran (2011).

4 Although SpVAR is less restrictive regarding the assumption of exogenous foreign variables, it is still subject to the assumption that the weight matrix should be exogenously identified. Therefore, this paper uses lagged trade flow and/or bank claim stocks to construct the weights. As trade flow and/or bank exposure in the previous periods are less likely to be affected by the default in future, it is argued that the weight matrix can be regarded as exogenous to the dependent variables. The estimation of SpVAR with endogenous weights is left for future research.

5 Given the fact that the estimation of the spatial vector error correction model (SpVECM) under the assumption of ‘endogenous foreign variables’ is still at a very preliminary stage, extending the SpVAR model to the SpVECM model is left for future research.

6 W is typically a square matrix in the spatial econometric literature. However, note that W in principle does not need to be square in size. Recently developed models, such as the mixed cross-section GVAR (Gross & Kok, 2013; Gross et al., 2016) and unbalanced spatial panel regression (Baltagi, Bresson, & Etienne, 2015), for example, do not require a square matrix.

7 An alternative definition could be the average amount of export from country h to country l over the past 12 months before period t (Eder & Keiler, 2015); or initial trade in January 2008 before the crisis period. Both definitions provide qualitatively robust results. This indicates that the assumption of ‘endogenous weights’ is likely to be held. Detailed results are available from the author upon request.

8 CDS changes are defined as log differences in credit spread, or percentages of basis points. The log difference in spreads can be viewed as the return on buying credit protection and thus reducing credit risk (e.g., Acharya & Steffen, 2015; Alter & Beyer, 2014; Alter & Schüler, 2013; Gross & Kok, 2013).

9 For a detailed comparison of in-sample predictive accuracy, see Appendix E in the supplemental data online. Detailed results are available from the author upon request.

10 Dees et al. (2007) use GDP as the weights for individual shocks. GDP is the average GDP for the 11 countries over the period 2008–13. The results based on GDP weights are completely robust. Detailed results are available from the author upon request.

11 Appendix D in the supplemental data online illustrates the response of the 11 countries to a shock to Italian sovereign, banking and corporate CDS, as an example. Note that the quick die out of the response might be due to the fact that the long-term relationship has not been integrated into the SpVAR model. When the ECM is incorporated, the response may last for a longer time. However, this shortage will not fundamentally change the conclusion. The lack of persistence of the response is consistent with previous findings (e.g., Chen et al., 2010; Gray et al., 2013).

12 Decomposition for each individual country is available from the author upon request.

13 For detailed definition, see Appendix F in the supplemental data online. EBA/ECB/SSM data were only published twice: once in 2011 and again in 2014. Therefore, the weights from 2008 to 2011 are based on the data published in 2011; the rest of the weights are based on the data in 2014.

14 The author gratefully acknowledges an anonymous reviewer for suggesting this approach.

15 For the results, see Appendices G and H in the supplemental data online.