Quality evaluations and the breakdown of statistical herding in the dynamics of box-office revenue

Are motion picture audiences influenced by box office reports? Do they `herd' after the leaders and ignore possibly better but less popular films? How important is a big opening to the revenue a film eventually earns? This paper develops a dynamical learning model of motion picture demand that reveals the influence of prior demand on current and future demand. The model shows that audience dynamics are complex and herding is fragile. We show that the path of box office revenue bifurcates into a hit and a non-hit branch four weeks into the run and these paths rapidly diverge thereafter.