Learning by suing: structural estimates of court errors in patent litigation

This paper presents structural estimates of the probability of validity, and the probability of Type I and Type II errors by courts in patent litigation. Patents are modeled as uncertain property rights, and implications of the model are tested using stock market reactions to patent litigation decisions. While court errors are inherently unobservable, the estimation quantifies beliefs about patent validity and court errors in a Bayesian context by relying on observable win rates and stock market reactions. I estimate that the underlying beliefs about validity average from 0.55 to 0.70 for litigated patents. For a number of different specifications, I show that Type I errors (finding a valid patent invalid) occur with an estimated probability of 0.20 to 0.25. The range for Type II errors (finding an invalid patent valid) varies more broadly, from near zero probability to as high as 0.40. Additional implications of the model address patent value