The authors discuss the problems with the “in-sample prediction” method.
In several recent antitrust class certification cases, plaintiffs have used a novel econometric method, referred to as “in-sample prediction,” to claim empirical evidence of class-wide antitrust impact. Proponents of this method claim that it can establish impact for individual observations (i.e., transactions). They identify an at-issue transaction as impacted when the actual price is more than the predicted but-for price.
While several courts have accepted this method to date, coauthors Nikhil Gupta and Matthias Lux argue in this article that in-sample prediction is not reliable. By carefully analyzing the way this method predicts individual impact, they show that, contrary to claims of the method’s proponents, it does not predict the causal impact of the challenged conduct on a specific transaction.
This article was originally published in the Antitrust Source in November 2025.
Boston 
