Performance and Resilience of Collaboration Networks
Duration 12 months (March 2015 - February 2016)
Funding source ETH Zurich Foundation, through ETH Risk Center
The goal of this project is to understand systemic risk in collaboration networks. We want to predict how the failure of one or a few agents – or links – can hamper the performance of the network. Consequently, we intend to define a quantitative risk measure (resilience), that can be validated and tested on a set of real collaboration networks.
We argue that there exist two types of collaboration networks: one in which agents have aligned utilities, such as Open Source Software (OSS) communities, where developers contribute to the same common goal; and one in which agents have misaligned utilities, such as inter-firm R&D networks or co-autorship networks, where firms or scientists collaborate in a competitive environment to increase their individual pay-off, without being directly concerned for a collective benefit.
In this project, we want to provide a quantitative definition of network performance for both cases. A logical consequence of this definition will be the quantification of how such performance changes when one or few agents – or links – are removed or suffer a shock, either at the aggregate or at the individual level, depending on the nature of the collaboration network under examination.
Figure: emergence and decline of a giant connected component in the computer software R&D network (adapted from Tomasello et al., 2013, "The Rise and Fall of R&D Networks")
In addition, we want to define a new temporal risk measure for collaboration networks, that is not uniquely topology-based and that is able to correctly quantify the change in aggregate or individual performance, when one of the agents (or links) suffers a shock or is forcedly removed from the collaboration network. Following the definition of such a measure, we want to investigate whether real collaboration networks are resilient and, in case they are not, whether it is possible to overcome these sub-optimalities and design risk-optimized collaboration networks. The research questions of interest can be summarized as follows: