Modelling industry interdependency dynamics in a network context

Purpose - Interdependency among industries is vital for understanding economic structures and managing industrial portfolios. However, it is hard to precisely model the interconnecting structure among industries. One of the reasons is that the interdependencies show a different pattern in tail events. This paper aims to investigate industry interdependency with the tail events. Design/methodology/approach - General predictive model of Rapachet al.(2016) is extended to an interdependency model via least absolute shrinkage and selection operator quantile regression and network analysis. A dynamic network approach was applied on the Fama–French industry portfolios to study the time-varying interdependencies. Findings - A denser network with heterogeneous central industries is found in tail cases. Significant interdependency varieties across time are shown under dynamic network analysis. Market volatility is identified as an influential factor of industry connectedness as well as clustering tendency under both normal and tail cases. Moreover, combining dynamic network with prediction direction information into out-of-sample industry return forecasting, a lower tail case is obtained, which gives the most accurate prediction of one-month forward returns. Finally, the Sharpe ratio criterion prefers high-centrality portfolios when tail risks are considered. Originality/value - This study examines the industry portfolio interactions under the framework of network analysis and also takes into consideration tail risks. The combination of economic interpretation and statistical methodology helps in having a clear investigation of industry interdependency. Moreover, a new trading strategy based on network centrality seems profitable in our data sample.