A Graph Learning Model of Network Resources for Early Stage Startup Success Prediction

With high profitability accompanied by high risks, startups are driving industry evolution and innovation. However, due to high information asymmetry, startup success prediction remains challenging. From the perspective of network resources, we propose a variant heterogeneous graph attention network (ResourceNet) to model how a focal startup can access and leverage network resources from inter-organizational networks connected by investment, co-portfolio, and VC syndication relationships, for future success. We follow the design science paradigm to develop the node and link-aware attention mechanisms in graph network representations that jointly explore the impact of different mechanisms explaining the value of network resources, i.e., reach, richness, and receptivity. This project provides contributions to the startup success prediction studies by demonstrating the value of network resources in a topological interorganizational network, and also important managerial implications for startup companies (to seek network resources for future success) and VC (to pick the winners).