Dynamic Network Analysis Via Exponential Random Graph Models

Networks consist of nodes and edges,where nodes represent individuals or organizations and edges represent relationships among nodes.Networks are rep-resentations of relational data.We can use both network indices and models to analyse real data.Exponential random graph model is widely used in mod-elling networks,which assumes the network is decided by local configurations.In reality,we are exposed to an increasing number of dynamic networks,where nodes and edges are appearing and disappearing continuously.It is important to model the evolution of dynamic networks.In this thesis,we propose a dynam-ic model which is capable of estimating different parameters at the same time.The model has more accurate estimations to networks and it makes full use of data from adjacent similar networks.The model is estimated through iteratively reweighted least squares approach.Simulation results also show that our model is helpful in detecting changing trends,comparing parameters,and modeling the evolution of networks.International trade data can be analyzed using network methods.Specifi-cally,the analysis of network features is helpful in understanding the size of the trade network and the role of each country.Dynamic exponential random graph models are useful to understand the changing impact of factors on trade.We built a collection of 16 networks for 60 countries with Gross Domestic Product(GDP)and distance as covariates.The analysis reveals the following:between the years 2001 and 2016 the size of trade network increased;the USA and Ger-many played key roles in the trade market:China developed more trade partners and was one of the fastest growing economies of this period:countries were more inclined to have bilateral trade especially in boom years:GDP had a more vari-able impact on export than on import;distance had a more consistent impact on trade.