How BAT Predictor Works
BAT Predictor uses a series of time-sensitive artificial intelligence models to predict final auction prices based on 50,000+ historical auctions, 1,200,000+ historical bids, and 3,700,000+ comments.
Our machine learning models understand that as an auction progresses, we have more information about how things might end up. Throughout the auction, our models get increasingly certain about the outcome, while carefully showing the bounds of it's certainty. During every auction, 98% of our confidence bands include the final auction price, and the most likely value presented goes from being within 31% of the final price to 8% of the final price. Below are some out-of-sample accuracy reports and series of scatterplots that demonstrate how certainty tightens up over 12 hour increments. We pool estimates from all our models over the course of an auction to give the most up-to-minute predictions - and give you an informational advantage over other bidders. Read more about the architecture behind the BAT Predictor.