Analysis and Prediction of Venture Capital Network

Abstract

Venture capital network reveals co-investment relationships between investors. The comprehensive analysis of its structure could not only detect the investment preference of each investor, but also uncover some hidden patterns within their cooperation's engagement. In this paper, we construct two networks-one directed and another undirected-with nodes as investors and edges as their collaboration relationships to analyze the investor networks from different perspectives. We first analyze two networks based on conventional techniques from graph theory such as community detection and PageRank centrality and obtain some interesting findings concerning the patterns of decision-making and each investor's prestige. Given such indicative discovery, we also attempt to predict investors' characteristics with Graph Convolutional Network (GCN) and Personalized Propagation of Neural Predictions (PPNP) by encoding each investor's information into a vector embedding. The results once again display the power and advantages of these two emerging neural networks, and it would also facilitate further research and analysis.