Modelling the resilience and disruption propagation in business networks at the firm level: Introduction of previous research and future plans (original) (raw)

Petr Matous Philippa Pattison Ian Wilkinson Pascal Perez Rob Axtell Yasuyuki Todo Peng Wang National economies comprise complex networks of firms exchanging material and non-material resources via physical infrastructure, which can be modelled by agent-based models (ABMs). The network structure of interfirm interactions effects how shocks and fluctuations in one part propagate, disrupt behaviour and affect aggregate outcomes and performance. The critical importance of the structure and behaviour of these networks is being increasingly recognised but, to date, sufficiently realistic models have not been developed of these highly complex systems, due to lack of data (complete interfirm production network data for an entire economy are available only in Japan) and the ability to adequately model the firms’ behaviour. A correct specification of interaction network topologies is of crucial importance in ABM but previous models have been operated mainly on theoretically stylized networks such as lattices, random networks, or generic small-world and scale-free networks. ABMs have been mainly used to illustrate theoretical principles rather than being explicitly calibrated against statistical data from real-world networks. Exponential Random Graph Models (ERGMs) support statistical inference regarding the processes underlying observed network structures, efficiently capturing complex network “motifs”, which can represent, for example, the tendencies to form cliques in trade networks or utilizing intermediaries who broker between diverse network cliques. Moreover, multi-level ERGMs, introduced by Wang et al (2013), can be used to explore complex interactions between supply networks and underlying physical infrastructure networks. Importantly, ERGM coefficients can be obtained from adequately sampled network data. This project develops ABM embedded in empirically estimated multi-level network structures governed by interaction rules identified in Australian business networks. The model is developed by (1) collecting snowball samples of supply networks in diverse regions of Australia and integrating them with underlying physical infrastructure network data; (2) estimating the micro structure of these multilevel network fragments by ERGMs; (3) setting up a 2 million-agent ABM with contact structures probabilistically inferred in the previous step; (4) validating and calibrating the process on the existing data of the entire production network in Japan. Such model can simulate fine-grained non-linear diffusion processes and their pathways, which could not be accomplished in the ERGM framework alone. The aim of model is to be used by policymakers and firms to examine the critical infrastructure for the resilience of the system, the effects of different types of shocks on the system, and to develop strategies to deal with them in effective ways.