Impact of Social Network on Financial Decisions

Abstract

Using a financial data set with attribute vectors, we simulate a network graph based on the mutual distances of the attribute vectors of the individuals. We learn that a network graph simulated via means of distance computation is most likely to be transitive. Running a triad census in our actual network graph from Facebook, we find that the number of transitive triads is much more than what is to be expected from chance. So, we guessed at whether this transitivity is from attribute similarity. To verify our guess, we focus on the attribute information of network data of Facebook. We construct the attribute vectors of the individuals by defining a similar metric as in the case of financial data, we simulate the graph of the attribute information of the individuals. We saw that a considerable percentage of the edges of the actual network graph is being predicted by the simulated graph, although the simulated graph grossly overestimates the total number of edges. As attributes play roles in network formation, it is likely that the network parameters will add information and improve the regression on the financial data set. So, we run the regression with both types of variables and find that eigenvector centrality and the clustering coefficient indeed improve the regression results as additional regressors.