Abstract: In this paper, we describe research and an application of agent based modeling to create financial network data. Creating a dataset of this type presented some unique challenges. First, the dataset we are trying to emulate is large and sparsely connected (20 million nodes, 20 million edges, in 500GB). Second, it includes multiple types of entities and relationships. A system made up of multiple types of entities with various relationships is tailor made for agent based modeling. Third, this dataset is being created as part of a larger project that is creating graph analysis tools that will work with massive, dynamic datasets. Therefore, it is important that we be able to control what the generated dataset contains so we can test various parts of our graph analysis system. An initial agent based model has been created using Netlogo. This prototype is being created iteratively as we continue to investigate the patterns and other features within the actual dataset. The domain in which the graph analysis tools are to be used is, understandably, of a sensitive nature. We wish to keep the datasets we produce unclassified so they can be released to the academic and analytic communities to aid in collaboration. This presents its own challenges as we need to produce a dataset that is a reasonable facsimile of the actual data for meaningful collaboration, however not so similar as to represent any sort of unreasonable disclosure of information.