Financial connectedness revisited: the role of Fama-French risk factors

ABSTRACT

We study the contributions of Fama-French type risk factors to spillovers across global stock markets by combining the framework proposed by Diebold and Yilmaz(2009) with insights from asset pricing. We demonstrate that incorporating the risk factors absorbs approximately 40% of information from DY’s total spillover measure to blur the boundaries between ‘To’ and ‘From’ countries and alleviates its upward trend. We find that the DY’s spillover index after controlling for the risk factors yet fluctuates in accordance with historically important economic events over time. Last but not least, the stock market characteristics implied by risk factor exposures are revealed to play the crucial role in determining the Net spillover direction among global equity markets.

KEYWORDS:

• Financial connectedness
• global equity markets
• risk factors
• generalized forecasting error decomposition

JEL CLASSIFICATION:

• C32
• G12
• G15

Disclosure statement

No potential conflict of interest was reported by the authors.

SupplementarySupplementary materialSupplementary material

Supplementary material can be accessed for this article here.

Notes

¹ Two key assumptions Campbell and Viceira (1999, 2002) and Campbell, Chan, and Viceira (2003)are i) the expected excess return on a risky asset follows an AR(1) process, and ii) the dynamics of relevant state variables follow a VAR(1) process.

² For more detailed economic description of the rolling-sample Total connectedness, see Diebold and Yilmaz (2015).

³ Forbes and Rigobon (2002) define financial contagion as a significant increase in cross-market linkage after a shock to one country.

⁴ We also try FF 5-factors including CMA and RMW to see if our results are robust to different factor structure. As a result, both the estimated factor loadings of the market, size and value factors and the full-sample connectedness are very similar to our results with FF 3-factors. Empirical results with the 5-factors are available in the online appendix.

⁵ We have verified whether our results would change if we consider Pástor and Stambaugh (2003)’s market liquidity factor, which is considered important in the literature such as Adrian et al. (2017). Contrary to Racicot and Rentz (2017), we find no statistical significance of the liquidity factor. The result from the fact that the market liquidity factor is constructed based on the US stock market whereas our analysis covers country stock market indices. The estimation results are available in the online appendix.

⁶ Note that our model specification is different from equation (7) in Petkova (2006) where the author employed FF factors as regressors. This question would be interesting and left for future research.

⁷ We exclude JPN in our panel regression to avoid its well-recognized anomalous behaviours.

⁸ Since equation (8) includes constructed regressors where measurement errors are suspected, we applied the GMM-based instrumental variable technique proposed by Racicot and Rentz (2017) for robustness check. Although the estimated results of Panel A are mostly similar, the difference of coefficients between ‘To’ and ‘From’ groups in Panel B becomes less clear. The detailed results of the GMM estimation are available in the online appendix.