The financial markets have been affected by the recent advancements in social media. Retail traders often use social media as a source of financial information. This thesis provides a mathematical proof of concept for how retail traders on social media can cause a large change in market price and that it is possible to identify the factors that drive the price change before it is seen in the market. An Agent Based Model is used to simulate a self-contained artificial market on a network. Traders gather pricing information from social media posts created by traders on the network and their own pricing techniques. Trader demographics and network structures are varied to study their results. A network with trader demographics of predominately momentum traders has a threshold corresponding to a large change in market price. The threshold, first discovered in this thesis, is denoted as the Herd Threshold. There is a negative relationship between network clustering and the magnitude of the price change. A convolutional neural network is used to predict trader demographics and network structure. The convolutional neural network has greater success when classifying trader demographics than network structure. This thesis is a foundation for multiple avenues of future work.
Library of Congress Subject Headings
Stock exchanges--Mathematical models; Speculation--Mathematical models; Deep learning (Machine learning); Convolutions (Mathematics); Neural networks (Computer science); Social media--Economic aspects
Applied and Computational Mathematics (MS)
Department, Program, or Center
School of Mathematical Sciences (COS)
Weir, Benjamin, "Using an Agent Based Model and Deep Learning to Simulate and Infer Market Behavior on Networks" (2022). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus