Communications of the IIMA


This paper describes a forecasting competition developed for a graduate course in information decision management to illustrate a data mining technique. The competition was inspired by the Netflix Prize competition (Bennett & Lanning, 2007) and other crowdsourcing competitions. A dataset of 276 monthly observations from 1989 through 2011 was partitioned into a learning dataset of 264 observations and a holdout sample of the 12 observations for 2011. After being subjected to a teaching module covering time series forecasting methods, teams of students tried their luck forecasting the holdout sample with an objective of minimizing the root mean square error. We learned that combining models results in a 7% decrease in RMSE.