Joint tests of contagion with applications

Abstract

Joint tests of contagion are derived which are designed to have power where contagion operates simultaneously through coskewness, cokurtosis and covolatility. Finite sample properties of the new tests are evaluated and compared with existing tests of contagion that focus on a single channel. Applying the tests to daily euro zone equity returns from 2005 to 2014 shows that contagion operated mainly through higher order moment channels during the GFC and the European debt crisis, which were not necessarily detected by traditional tests based on correlations. The empirical results have important implications for pricing risk and constructing well diversified portfolios.

Keywords:

• Coskewness
• Cokurtosis
• Covolatility
• Lagrange multiplier tests
• European financial crisis
• Equity markets

JEL Classification:

• C1
• F3

Acknowledgments

We thank an anonymous referee, Joshua Chan, Thomas Flavin, Xun Lu, James Morley, Pengfei Wang, Benjamin Wong, Sen Xue, Wenying Yao, and participants of the brownbag seminar at Hong Kong University of Science and Technology, the National Taiwan University, the 20th International Conference of Computing in Economics and Finance, the Australasian Meeting of the Econometric Society and the Macro-Finance workshop at Deakin University.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental data

Supplemental data for this article can be accessed at https://doi.org/10.1080/14697688.2018.1475747.

ORCID

Renée Fry-McKibbin  http://orcid.org/0000-0003-2396-7540

Notes

1 A natural extension of the proposed joint tests is to construct a test focussing on the even-ordered moments of cokurtosis and covolatility. This test is not constructed here but could be done following the strategy of the joint test COSKEW. Instead, the approach adopted here is to interpret jointly the two joint tests JOINT and COSKEW and infer the relative role of higher order even moment channels. This strategy is further complemented by also interpreting the properties of the joint tests together with the single contagion channel tests.

2 These countries are Cyprus (joining 2008), Estonia (2011), Latvia (2014), Lithuania (2015), Malta (2008), Slovakia (2009) and Slovenia (2007). To minimize the volume of results reported in the empirical illustration of the joint test, Finland and Luxembourg are also excluded as they are not the main crisis affected countries.

3 Narayan and Sharma (2015) and Narayan et al. (2013) show that hypothesis tests in finance depend of data frequency. Daily data rather than weekly or monthly data is used in this paper as the analysis is conducted from the viewpoint of an investor seeking to diversify their portfolio as discussed in Section 2. The assumption of daily diversification underlies the analysis. As crises are fast moving and their transmission across markets is likely to be more evident at a daily frequency, daily data is the most appropriate measure for the contagion tests. Bannigidadmath and Narayan (2016) show that daily data improves upon the use of monthly data when the objective is to obtain as much information as possible from data as is the case here. Beginning with the seminal work of Forbes and Rigobon (2002), models of contagion from a financial market perspective are generally based on daily data. An alternative is to use a regime switching model such as in Boyer et al. (2006), Rodriguez (2007) where a switching-parameter copula is specified; or Chan et al. (2018) who specify a regime-switching skew-normal model.

4 The data source is Datastream. The mnemonics are: Austria—Austrian Traded index (ATXINDEX); Belgium—BEL 20 price index (BGBEL20); France—France CAC 40 price index (FRCAC40); Germany—MDAX Frankfurt price index (MDAXIDX); Greece—Athex composite price index (GRAGENL); Italy—FTSE MIB price index (FTSEMIB); Ireland—Ireland Se Overall price index (ISEQUIT); The Netherlands—AEX price index (AMSTEOE); Portugal—Portual PSI All Share price index (POPSIGN); Spain—IBEX 35 price index (IBEX351); US—Dow Jones Industrials price index (DJINDUS).

5 Time zones are accounted for by adjusting the returns using a two day rolling average of each return. Market fundamentals are accounted for by using the residuals of a VAR(5) estimated for all markets over the sample period (Forbes and Rigobon 2002, Fry et al. 2010, Fry-McKibbin and Hsiao 2018).

6 An alternative strategy to model changes in contagion over time is to follow Diebold and Yilmaz (2009, 2014) and perform the analysis over a rolling window.

7 In choosing the benchmark country within the euro zone to test for regional contagion, an alternative strategy would be to choose from the euro zone countries under financial stress from not being able to refinance their government debt for example.

Additional information

Funding

The authors gratefully acknowledge Australian Research Council ARC Discovery Project DP160102350 and DP120103443 funding and Macau University of Science and Technology Foundation for Faculty Research Grant FRG-17-019-MSB.