Computers Can’t Predict Every Deviation

We don’t do traditional screens and are actually looking for situations in which the publicly available information that a computer can analyze is giving false signals. For example, we bought auto-insurer Progressive [PGR] in 1999 when their earnings looked bad because they were spending heavily on a direct-to-consumer strategy like Geico’s. A computer would see that as a negative earnings trend resulting in a too-high multiple, but it doesn’t know how to judge whether certain spending might generate big returns in the future.