Magic: The Gathering (MtG) represents a significant and dynamic market for collectible trading cards, characterized by fluctuating prices driven by tournament results, player demand, and card rarity. This thesis explores time series forecasting techniques applied to the MtG card market, focusing on forecasting card prices using statistical and machine learning models. Specifically, the research compares the performance of traditional methods such as ARIMA, Random Walk, and NNETAR models to a proposed forecast combination neural network model.
A comprehensive database was created, combining price, tournament, and card attribute data, and feature engineering was employed to enhance the predictive power of the models. The methodology incorporates advanced statistical techniques and machine learning to build a more accurate and robust forecasting system. The results indicate that the proposed neural network model outperforms traditional methods in forecasting accuracy. This project also presents the ts.shiny application, an interactive tool which offers an accessible platform for visualizing and analyzing time series data.
The research concludes with insights into the factors driving MtG card prices and suggestions for improving forecasting models and applications in the future.