Predicting Steam Community Market Prices Using Linear and Non-Linear Models

Abstract

The Steam Community Market represents an increasingly complex digital market where users participate in the buying and selling of in-game items. Therefore, understanding how the Steam Community Market works may produce insights into the functioning of other digital markets that exist in areas not only in the world of gaming, as well as elucidate the psychology of a gamer or participant in such market. This thesis aims to shine some light on the inner workings of the Steam Community Market by tackling the question of how items are priced on the Steam Community Market. Specifically though the lens of Counter-Strike 2 and its items, linear (OLS, Ridge, LASSO, Elastic Net) and non-linear (Polynomial, Kernel Ridge) models are used to better understand the modeling and predicting of item pricing based on their inherent features. Through the process of modeling and predicting, the relationship between features and price prove to be more complex than a simple linear relationship. We find that modeling these relationships of item pricing based on features give insight into what features are valued more than others, which can help model prices of future items released onto the Steam Community Market as well as price current items’ sensitivity to different stimuli.