Modeling movie success when "nobody knows anything": Conditional stable distribution analysis of film returns

In this paper we apply a recently-developed statistical model that explicitly accounts for the extreme uncertainty surrounding film returns. The conditional distribution of box-office returns is analyzed using the stable distribution regression model. The regression coefficients in this model represent what is known about the correlates of film success while at the same time permitting the variance of film success at the box office to be infinite. The empirical analysis shows that the conditional distribution of film returns has infinite variance, and this invalidates statistical inferences from the often-applied least-squares regression model. The estimates of the stable regression confirm some earlier results on the statistics of the movie business and the analysis demonstrates how to model box-office success in the movie business where ``nobody knows anything''