Measuring financial contagion using general social interaction model with trade network structure

Abstract

This article applies general social interaction model to measure contagion across stock markets in a given trade network. We divide the whole sample period into five types of subperiods according to the extent of negative and positive coexceedances. We find that interdependence and contagion in the given trade network exist significantly. Our empirical results also confirm previous evidences that contagion is stronger for extreme negative returns than for positive ones. However, the direction of asymmetry reverses when the degree of coexceedances becomes somewhat smaller. In addition, trade partners’ macroeconomic fundamentals in trade network significantly impact on domestic stock market during extreme periods.

Keywords:

• contagion
• general social interaction model
• trade network
• interdependence

JEL Classification:

• C33
• F15

Funding

The research is supported by the Natural Science Foundation of China [grant number 71171119] and Humanities and Social Sciences Program of China’s Ministry of Education [grant number 12JJD790017].

Notes

¹ As there are disagreements on how to define contagion in the academic literatures, this article clarifies definitions of interdependence and contagion suggested by Forbes (2012): interdependence is high correlations across markets during all states of the world and contagion is the spillovers from extreme negative events (put another way, markets move more closely together during periods of crisis).We also see the spillovers from extreme positive events as an alternative form of contagion.

² See Section II.

³ See also Section II.

⁴ The feedback and indirect effects of contagion are illustrated by the case of Global Financial Crisis of 2008. The US subprime crisis impacts France and UK. Feedback effect of contagion refers to the impact of France on US and indirect effect of contagion refers to the impact of France on UK.

⁵ See also Section II.

⁶ Since this article does not divide the N countries into more than one group, the unobserved group-specific fixed effects (Lee, 2007) degenerate into the intercept.

⁷ It is standard to require that  < 1, which is realistic that the performance of one country’s market cannot be totally influenced by its trade partners’ markets.

⁸ The estimation approaches of general social interaction model are usually inspired by spatial econometric literatures. Ordinary least squares estimation is no longer consistent here. Kapoor et al. (2007) uses a generalization of the generalized moment estimator suggested in Kelejian and Prucha (1999) to estimate disturbances only. Mutl and Pfaffermayr (2011) extend their work for estimating spatial autoregressive parameter as well.

⁹ We choose only 17 countries to ensure data integrality and quality, for package splm only focuses on balanced panels.

¹⁰ Using interquartile range to identify outliers considers the difference in the degree of dispersion among return distributions, whereas the 5th (95th) quantile treats them alike.