Credit risk-free sovereign bonds under Solvency II: a cointegration analysis with consistently estimated structural breaks

Abstract

European insurance and reinsurance undertakings are facing the advent of a new regulatory framework. In the current proposal for its technical specifications under the pillar 1 standard formula, sovereign debt of European Union (EU) member states is treated as risk-free. This article examines the validity of this assumption for 26 EU member states. Taking into account the possibility of multiple structural breaks, we find evidence for the convergence of government bond yields of several countries with the yields of a risk-free asset. For the majority of countries, however, there is no such evidence. A detailed discussion of regime shifts in relation to European bond market integration is provided. Our findings have important implications for insurance companies, bond investors and regulators alike.

Keywords:

• European bond markets
• cointegration
• structural breaks
• Solvency II
• financial crisis

JEL Classification:

• C51
• F36
• G15
• G18

Acknowledgements

I would like to thank the anonymous reviewers for helpful comments which improved the quality of this paper. All remaining errors are my own.

Funding

This work was supported by the PhD programme of zeb/rolfes.schierenbeck.associates.

Notes

¹ Sovereign debt is a key investment class for insurance companies (BIS, 2011; Swarup, 2012). Insurance Europe and Oliver Wyman (2013) report a value of EUR 2.4 trillion for government bonds held by insurance companies per 31 December 2011.

² The majority of calculations in this article were performed using gretl. To simulate critical values for the cointegration test of Arai and Kurozumi (2007) and Kejriwal (2008), we used Excel-VBA. All program code are available from the author upon request.

³ The maximum number of breaks M that can be accounted for is a result of the choice of the trimming parameter . All partitions [T_j−1 + 1, T_j] need to be at least of length . A standard setting in the literature is .

⁴ We present the UD_max and SEQ(k + 1|k) test statistics in the notation of Kejriwal (2008).

⁵ For reasons of data quality (see Section V), the interest rate series of Luxembourg was excluded from this analysis.

⁶ All data fulfil the selection guidelines related to the Maastricht Treaty convergence criteria as summarized before with the following exceptions: primary market yields are used for Cyprus (whole sample), Bulgaria and Romania (both till December 2005), Slovenia (till October 2003) and Lithuania (till October 2007). As there was no outstanding long-term sovereign debt for Luxembourg till May 2010, long-term bonds issued by a private credit institution were used for the time before May 2010 instead.

⁷ Test results for the relation Poland vis-à-vis the Czech Republic were derived using Johansen’s (Johansen, 1988) standard trace test and are spared here to save space. Details are available from the author upon request.

⁸ Table 3 reports a convergence of coefficients from large to values around for these countries that support the argument for closer ties of bond yields after the introduction of the Euro.

⁹ According to Lund (1999), before the decision of the European Council in May 1998, uncertainty remained regarding the selection of member countries and a possible postponing of the Monetary Union.

¹⁰ Camarero et al. (2002) even find long-run convergence between Austria, Germany, the Netherlands and Luxembourg with the Maastricht criterion for interest rate convergence for the period 1980 to 1996. Ireland and Belgium were still in the process of catching-up in the period 1980 to 1996, which is not a contradiction to our results. The findings obtained by Lund (1999) are also consistent with our results as he showed that market-implied EMU membership probabilities for the Benelux countries were already 100% for the period 1995 to 1998, which results from narrow yield spreads with Germany already before the introduction of the Euro. Austria and Ireland were not considered in his analysis.

¹¹ After the regime shift, there is still long-run convergence between Germany and the Netherlands as for the Netherlands (see Table 3). The break obtained from the test occurs in the mean spread between these countries that rises from 6 basis points to 36 basis points after the break. For Austria and France, however, divergence from German yields is indicated by delta coefficients significantly lesser than 1 (see Table 3).