Using past contribution patterns to forecast fundraising outcomes in crowdfunding

The crowdfunding mechanism has proven to be a practical way of raising funds, especially with the widespread use of the Internet. However, one limitation of current crowdfunding platforms is that it is hard for creators and backers to anticipate the success of a campaign. This paper tackles this limitation. We take a two-pronged approach to building our forecasting model. First, we explore the nature and heterogeneity of contribution dynamics in crowdfunding campaigns and compare them across two natural groups (successful and unsuccessful campaigns). We then use insights generated from our exploratory analysis and draw upon the general laws of motion for stochastic processes in order to introduce a new dynamic model for predicting crowdfunding outcomes. Our model incorporates the history and dynamics of both the focal crowdfunding campaign and other campaigns for predicting outcomes. We compare our model to other parametric and semi-parametric benchmark models, and show substantial improvements.